Description:FORM 2

THE PATENTS ACT, 1970
(39 of 1970)
&
THE PATENT RULES, 2003

COMPLETE SPECIFICATION
(See Section 10 and Rule 13)

Title of invention:
SUPERVISED CONTRASTIVE LEARNING BASED MULTISENSOR PROGNOSTICS TO ESTIMATE REMAINING USEFUL LIFE OF A MACHINE

Applicant

Tata Consultancy Services Limited
A company Incorporated in India under the Companies Act, 1956
Having address:
Nirmal Building, 9th floor,
Nariman point, Mumbai 400021,
Maharashtra, India

Preamble to the description:

The following specification particularly describes the invention and the manner in which it is to be performed.
TECHNICAL FIELD
The disclosure herein generally relates to the field of machine prognosis, and more specifically to methods and systems for supervised contrastive learning based multi-sensor prognostics to estimate remaining useful life (RUL) of a machine.

BACKGROUND
Estimating a remaining useful life (RUL) of the machine used in an industry such as a manufacturing equipment or non- industry such as a vehicle engine, and the prognosis of the machine is very important to take precautionary measures from time to time. One way to estimate the remaining useful life (RUL) of the machine is based on the senser data obtained from the sensors present in the machine. Conventional machine learning (ML) or deep learning-based solutions (DL) for estimating the remaining useful life (RUL) of the machine using the temporal sensor data are very limited due to the following limitations. The first limitation is that each training sample instance must represent a full lifecycle of the similar types of machines (or previous runs of the same machine) i.e., each training instance must contain start to end of the machine lifecycle. But in the real world, there is hardly any properly annotated dataset that satisfy the full-lifecycle data of the machine.
The second limitation is that majority of the training sample instances are incomplete i.e., not representing the entire lifecycle from the start to the end but only partial lifecycles (e.g., mid-life to end etc.). Further, the sample instances are of different and varied lengths. e.g., for one instance 110 time-steps (i.e., data-points) from the last i.e., before the engine replacement point is given and for some other it is 70 time-steps, etc. Hence it is technically challenging to apply the ML or DL-based solutions for estimating the remaining useful life (RUL) of the machine using the temporal sensor data. Further, the ML or DL-based solutions for estimating the remaining useful life (RUL) of the machine are not efficient and accurate mainly due to lack of a proper uniform temporal sensor data.

SUMMARY
Embodiments of the present disclosure present technological improvements as solutions to one or more of the above-mentioned technical problems recognized by the inventors in conventional systems.
In an aspect, a processor-implemented method for supervised contrastive learning based multi-sensor prognostics to estimate remaining useful life of a machine is provided. The method include the steps of: receiving one or more training service instance datasets associated with a machine, wherein each training service instance dataset comprises a plurality of service instance records and wherein each service instance record is a time-series data and comprises one or more sensor values and a corresponding RUL value; extracting a training service instance sub-dataset for each of the one or more training service instance datasets, based on a predefined selection percentage value, wherein each training service instance sub-dataset comprises two or more service instance records out of the plurality of service instance records present in the corresponding training service instance dataset; forming one or more service instance record subsequences, for each training service instance sub-dataset, based on a predefined subsequence window size, wherein each service instance record subsequence comprises the two or more service instance records and a number of the two or more service instance records present in each service instance record subsequence is equal to the predefined subsequence window size; forming one or more time-series snippets for each service instance record subsequence of the one or more service instance record subsequences, to obtain a plurality of time-series snippets from the one or more training service instance datasets, using a predefined time-series snippet length, wherein each time-series snippet comprises the number of the service instance records equal to the predefined time-series snippet length; training a supervised contrastive regressive encoder, using the plurality of time-series snippets, to obtain a trained contrastive encoder; assigning a classification label from one of (i) a healthy status and (ii) an unhealthy status, for each service instance record of the plurality of service instance records present in each of the one or more training service instance datasets, using the predefined selection percentage value; obtaining one or more encoded features, for each service instance record of the plurality of service instance records present in each of the one or more training service instance datasets, by passing the corresponding one or more sensor values to the trained contrastive encoder; training a classifier, with the plurality of service instance records present in each of the one or more training service instance datasets, using the corresponding one or more encoded features and the corresponding classification label of each service instance record, to obtain a trained classifier; training a regressor, with the two or more service instance records present in each training service instance sub-dataset of the one or more training service instance datasets, using the corresponding one or more encoded features and the corresponding RUL value of each service instance, to obtain a trained regressor; receiving more test service instance records of a machine whose health to be monitored, wherein each test service instance record comprises one or more sensor test values; obtaining the one or more encoded features, for each test service instance record of the one or more test service instance records, by passing the corresponding one or more sensor test values to the trained contrastive encoder; passing the one or more encoded features obtained for each test service instance record of the one or more test service instance records, to the trained classifier, to predict one of: (i) the healthy status and (ii) the unhealthy status, of the machine; and passing the one or more encoded features obtained for each test service instance record of the one or more test service instance records, to the trained regressor, to predict the RUL value of the machine, if the machine is predicted as the unhealthy status during the classification.
In another aspect, a system for supervised contrastive learning based multi-sensor prognostics to estimate remaining useful life of a machine is provided. The system includes: a memory storing instructions; one or more Input/Output (I/O) interfaces; and one or more hardware processors coupled to the memory via the one or more I/O interfaces, wherein the one or more hardware processors are configured by the instructions to: receive one or more training service instance datasets associated with a machine, wherein each training service instance dataset comprises a plurality of service instance records and wherein each service instance record is a time-series data and comprises one or more sensor values and a corresponding RUL value; extract a training service instance sub-dataset for each of the one or more training service instance datasets, based on a predefined selection percentage value, wherein each training service instance sub-dataset comprises two or more service instance records out of the plurality of service instance records present in the corresponding training service instance dataset; form a one or more service instance record subsequences, for each training service instance sub-dataset, based on a predefined subsequence window size, wherein each service instance record subsequence comprises the two or more service instance records and a number of the two or more service instance records present in each service instance record subsequence is equal to the predefined subsequence window size; form one or more time-series snippets for each service instance record subsequence of the one or more service instance record subsequences, to obtain a plurality of time-series snippets from the one or more training service instance datasets, using a predefined time-series snippet length, wherein each time-series snippet comprises the number of the service instance records equal to the predefined time-series snippet length; train a supervised contrastive regressive encoder, using the plurality of time-series snippets, to obtain a trained contrastive encoder; assign a classification label from one of (i) a healthy status and (ii) an unhealthy status, for each service instance record of the plurality of service instance records present in each of the one or more training service instance datasets, using the predefined selection percentage value; obtain one or more encoded features, for each service instance record of the plurality of service instance records present in each of the one or more training service instance datasets, by passing the corresponding one or more sensor values to the trained contrastive encoder; train a classifier, with the plurality of service instance records present in each of the one or more training service instance datasets, using the corresponding one or more encoded features and the corresponding classification label of each service instance record, to obtain a trained classifier; train a regressor, with the two or more service instance records present in each training service instance sub-dataset of the one or more training service instance datasets, using the corresponding one or more encoded features and the corresponding RUL value of each service instance, to obtain a trained regressor; receive one or more test service instance records of a machine whose health to be monitored, wherein each test service instance record comprises one or more sensor test values; obtain the one or more encoded features, for each test service instance record of the one or more test service instance records, by passing the corresponding one or more sensor test values to the trained contrastive encoder; pass the one or more encoded features obtained for each test service instance record of the one or more test service instance records, to the trained classifier, to predict one of: (i) the healthy status and (ii) the unhealthy status, of the machine; and pass the one or more encoded features obtained for each test service instance record of the one or more test service instance records, to the trained regressor, to predict the RUL value of the machine, if the machine is predicted as the unhealthy status during the classification.
In yet another aspect, there is provided a computer program product comprising a non-transitory computer readable medium having a computer readable program embodied therein, wherein the computer readable program, when executed on a computing device, causes the computing device to: receive one or more training service instance datasets associated with a machine, wherein each training service instance dataset comprises a plurality of service instance records and wherein each service instance record is a time-series data and comprises one or more sensor values and a corresponding RUL value; extract a training service instance sub-dataset for each of the one or more training service instance datasets, based on a predefined selection percentage value, wherein each training service instance sub-dataset comprises two or more service instance records out of the plurality of service instance records present in the corresponding training service instance dataset; form a one or more service instance record subsequences, for each training service instance sub-dataset, based on a predefined subsequence window size, wherein each service instance record subsequence comprises the two or more service instance records and a number of the two or more service instance records present in each service instance record subsequence is equal to the predefined subsequence window size; form one or more time-series snippets for each service instance record subsequence of the one or more service instance record subsequences, to obtain a plurality of time-series snippets from the one or more training service instance datasets, using a predefined time-series snippet length, wherein each time-series snippet comprises the number of the service instance records equal to the predefined time-series snippet length; train a supervised contrastive regressive encoder, using the plurality of time-series snippets, to obtain a trained contrastive encoder; assign a classification label from one of (i) a healthy status and (ii) an unhealthy status, for each service instance record of the plurality of service instance records present in each of the one or more training service instance datasets, using the predefined selection percentage value; obtain one or more encoded features, for each service instance record of the plurality of service instance records present in each of the one or more training service instance datasets, by passing the corresponding one or more sensor values to the trained contrastive encoder; train a classifier, with the plurality of service instance records present in each of the one or more training service instance datasets, using the corresponding one or more encoded features and the corresponding classification label of each service instance record, to obtain a trained classifier; train a regressor, with the two or more service instance records present in each training service instance sub-dataset of the one or more training service instance datasets, using the corresponding one or more encoded features and the corresponding RUL value of each service instance, to obtain a trained regressor; receive one or more test service instance records of a machine whose health to be monitored, wherein each test service instance record comprises one or more sensor test values; obtain the one or more encoded features, for each test service instance record of the one or more test service instance records, by passing the corresponding one or more sensor test values to the trained contrastive encoder; pass the one or more encoded features obtained for each test service instance record of the one or more test service instance records, to the trained classifier, to predict one of: (i) the healthy status and (ii) the unhealthy status, of the machine; and pass the one or more encoded features obtained for each test service instance record of the one or more test service instance records, to the trained regressor, to predict the RUL value of the machine, if the machine is predicted as the unhealthy status during the classification.
In an embodiment, the one or more sensor values are obtained from one or more sensors present in the machine.
In an embodiment, each training service instance dataset associated with the machine, is indicative of servicing data of the machine, just before an abnormal condition is encountered.
In an embodiment, training the supervised contrastive regressive encoder, using the plurality of time-series snippets, to obtain the trained contrastive encoder, comprises: forming one or more training batches, from the plurality of time-series snippets, based on a predefined training batch size, wherein each training batch comprises a number of the time-series snippets equal to the predefined training batch size; and training the supervised contrastive regressive encoder with the time-series snippets present in each training batch, at a time, until the one or more training batches are completed to obtain the trained contrastive encoder, wherein training the supervised contrastive regressive encoder with the time-series snippets present in each training batch comprises: passing each time-series snippet present in the training batch, to the supervised contrastive regressive encoder, to obtain the one or more encoded features of the corresponding time-series snippet; calculating a loss function value for each time-series snippet, based on a similarity of the corresponding time-series snippet with each other time-series snippet present in the training batch, using the corresponding one or more encoded features; calculating a loss function value of the training batch, by adding the loss function value of each time-series snippet present in the training batch; and updating network parameters of the supervised contrastive regressive encoder, based on the loss function value of the training batch and a predetermined loss function threshold value.
In an embodiment, similarity of two time-series snippets is calculated based on a statistical RUL value associated with each time-series snippet, and wherein the statistical RUL value associated with each time-series snippet is obtained by computing a statistical characteristic among the RUL values of the service instance records present in the corresponding time-series snippet.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.

BRIEF DESCRIPTION OF THE DRAWINGS
The accompanying drawings, which are incorporated in and constitute a part of this disclosure, illustrate exemplary embodiments and, together with the description, serve to explain the disclosed principles:
FIG. 1 is an exemplary block diagram of a system for supervised contrastive learning based multi-sensor prognostics to estimate remaining useful life of a machine, in accordance with some embodiments of the present disclosure.
FIGS. 2A through 2C illustrates exemplary flow diagrams of a processor-implemented method for supervised contrastive learning based multi-sensor prognostics to estimate remaining useful life of a machine, in accordance with some embodiments of the present disclosure.
FIG. 2D illustrates an exemplary flow diagram for training a supervised contrastive regressive encoder, in accordance with some embodiments of the present disclosure.
FIG. 3A illustrates an exemplary flow diagram for obtaining a set of models to estimate remaining useful life of a machine, in accordance with some embodiments of the present disclosure.
FIG. 3B illustrates an exemplary flow diagram for estimating remaining useful life of a machine using a test data through the set of models, in accordance with some embodiments of the present disclosure.

DETAILED DESCRIPTION OF EMBODIMENTS
Exemplary embodiments are described with reference to the accompanying drawings. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. Wherever convenient, the same reference numbers are used throughout the drawings to refer to the same or like parts. While examples and features of disclosed principles are described herein, modifications, adaptations, and other implementations are possible without departing from the scope of the disclosed embodiments.
Conventional solutions that use multi-sensor timeseries data of varied lengths representing partial lifecycle (e.g., mid-life to end-life) are very limited to predicting the remaining useful life (RUL) of the machine. The visual difference of the multi-sensor timeseries data, plotted using any descriptive methodology such as histogram, of n units of time (e.g. days), (now onwards to be referred as first set of timeseries data in a high dimensional vector space), prior to failure and (n + k) same time units (to be referred as the second set of timeseries data in the same dimension of vector space) prior to failure are not much distinct in case of the partial lifecycle data. Here, n is a small integer tends to zero and k is relatively bigger than n but not as big as it was possible for the full lifecycle data.
Hence, there is a need to increase the distance between first (1st) and second (2nd) set of time-series data in an embedded vector space after applying some suitable transformations. The present disclosure solves the technical problems in the art by employing contrastive learning to increase the distance among the candidates belonging to the different classes in the embedded space to address supervised multiclass classification problem. The supervised multiclass classification classifies the machine with a healthy status or unhealthy status using the embedded space obtained by the contrastive learning. Further, a regression model is employed to estimate the RUL of the machine using the embedded space obtained by the contrastive learning if the machine is said to be with the unhealthy status.
Referring now to the drawings, and more particularly to FIG. 1 through FIG. 3B, where similar reference characters denote corresponding features consistently throughout the figures, there are shown preferred embodiments and these embodiments are described in the context of the following exemplary systems and/or methods.
FIG. 1 is an exemplary block diagram of a system 100 for supervised contrastive learning based multi-sensor prognostics to estimate remaining useful life of a machine, in accordance with some embodiments of the present disclosure. In an embodiment, the system 100 includes or is otherwise in communication with one or more hardware processors 104, communication interface device(s) or input/output (I/O) interface(s) 106, and one or more data storage devices or memory 102 operatively coupled to the one or more hardware processors 104. The one or more hardware processors 104, the memory 102, and the I/O interface(s) 106 may be coupled to a system bus 108 or a similar mechanism.
The I/O interface(s) 106 may include a variety of software and hardware interfaces, for example, a web interface, a graphical user interface, and the like. The I/O interface(s) 106 may include a variety of software and hardware interfaces, for example, interfaces for peripheral device(s), such as a keyboard, a mouse, an external memory, a plurality of sensor devices, a printer and the like. Further, the I/O interface(s) 106 may enable the system 100 to communicate with other devices, such as web servers and external databases.
The I/O interface(s) 106 can facilitate multiple communications within a wide variety of networks and protocol types, including wired networks, for example, local area network (LAN), cable, etc., and wireless networks, such as Wireless LAN (WLAN), cellular, or satellite. For the purpose, the I/O interface(s) 106 may include one or more ports for connecting a number of computing systems with one another or to another server computer. Further, the I/O interface(s) 106 may include one or more ports for connecting a number of devices to one another or to another server.
The one or more hardware processors 104 may be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuitries, and/or any devices that manipulate signals based on operational instructions. Among other capabilities, the one or more hardware processors 104 are configured to fetch and execute computer-readable instructions stored in the memory 102. In the context of the present disclosure, the expressions ‘processors’ and ‘hardware processors’ may be used interchangeably. In an embodiment, the system 100 can be implemented in a variety of computing systems, such as laptop computers, portable computers, notebooks, hand-held devices, workstations, mainframe computers, servers, a network cloud and the like.
The memory 102 may include any computer-readable medium known in the art including, for example, volatile memory, such as static random access memory (SRAM) and dynamic random access memory (DRAM), and/or non-volatile memory, such as read only memory (ROM), erasable programmable ROM, flash memories, hard disks, optical disks, and magnetic tapes. In an embodiment, the memory 102 includes a plurality of modules 102a and a repository 102b for storing data processed, received, and generated by one or more of the plurality of modules 102a. The plurality of modules 102a may include routines, programs, objects, components, data structures, and so on, which perform particular tasks or implement particular abstract data types.
The plurality of modules 102a may include programs or computer-readable instructions or coded instructions that supplement applications or functions performed by the system 100. The plurality of modules 102a may also be used as, signal processor(s), state machine(s), logic circuitries, and/or any other device or component that manipulates signals based on operational instructions. Further, the plurality of modules 102a can be used by hardware, by computer-readable instructions executed by the one or more hardware processors 104, or by a combination thereof. In an embodiment, the plurality of modules 102a can include various sub-modules (not shown in FIG. 1). Further, the memory 102 may include information pertaining to input(s)/output(s) of each step performed by the processor(s) 104 of the system 100 and methods of the present disclosure.
The repository 102b may include a database or a data engine. Further, the repository 102b amongst other things, may serve as a database or includes a plurality of databases for storing the data that is processed, received, or generated as a result of the execution of the plurality of modules 102a. Although the repository 102b is shown internal to the system 100, it will be noted that, in alternate embodiments, the repository 102b can also be implemented external to the system 100, where the repository 102b may be stored within an external database (not shown in FIG. 1) communicatively coupled to the system 100. The data contained within such external database may be periodically updated. For example, data may be added into the external database and/or existing data may be modified and/or non-useful data may be deleted from the external database. In one example, the data may be stored in an external system, such as a Lightweight Directory Access Protocol (LDAP) directory and a Relational Database Management System (RDBMS). In another embodiment, the data stored in the repository 102b may be distributed between the system 100 and the external database.
Referring to FIGS. 2A through 2C, components and functionalities of the system 100 are described in accordance with an example embodiment of the present disclosure. For example, FIGS. 2A through 2C illustrates exemplary flow diagrams of a processor-implemented method 200 for supervised contrastive learning based multi-sensor prognostics to estimate remaining useful life of a machine, in accordance with some embodiments of the present disclosure. Although steps of the method 200 including process steps, method steps, techniques or the like may be described in a sequential order, such processes, methods and techniques may be configured to work in alternate orders. In other words, any sequence or order of steps that may be described does not necessarily indicate a requirement that the steps be performed in that order. The steps of processes described herein may be performed in any practical order. Further, some steps may be performed simultaneously, or some steps may be performed alone or independently.
At step 202 of the method 200, the one or more hardware processors 104 of the system 100 are configured to receive one or more training service instance datasets associated with a machine. Each training service instance dataset of the one or more training service instance datasets includes a plurality of service instance records. Each service instance record is in the form of a time-series data and includes one or more sensor values and a corresponding RUL value. Further, each service instance record includes a time-step which is a time instance on which the one or more sensor values and the corresponding RUL value are acquired or annotated. The RUL value of the machine is indicative of how many days or hours left for healthy working of the machine. Hence, the RUL value of the machine is measured in time, for example, number of days, number of hours, and so on, and measurement is considered based on the criticality of healthy working of the machine. Based on the RUL value of the machine, appropriate measures can be taken before encountering abnormality or non-working of the machine.
Each training service instance dataset represents a partial lifecycle data of the machine. In an embodiment, a number of the plurality of service instance records (denoted as a length of each training service instance dataset) present in each training service Instance dataset need not be same. So, each training service instance dataset associated with the machine, is a historical data indicative of servicing data of the machine, just before an abnormal condition is encountered or the machine whose RUL value is near to zero or zero, or just before the maintenance has happened. The training data having the both the healthy data and the abnormality data is very useful to train a machine learning model. In an embodiment, each training service instance dataset is identified with a unique instance identification number (ID). The one or more training service instance datasets is associated with a single machine or associated with multiple machines of similar type. In an embodiment, the one or more training service instance datasets is stored in the repository 102b of the system 100.
In an embodiment, the machine includes an apparatus, an equipment, a system, or any other machine whose healthy status is to be closely monitored or maintenance is required, or the prognosis to be measured at regular time intervals. In an embodiment, the machine includes any machine used in the industry or non-industry, including but are not limited to a manufacturing industry, an energy industry, a utility industry, vehicle, and a vehicle engine.
In an embodiment, the one or more sensor values are obtained from one or more sensors present in the machine. The one or more sensors present in the machine are used to measure the characteristics, functions, properties associated with the machine. The type of sensors present in each machine may be same or different, and in the same or variable number. In an embodiment, the type of sensors includes but are not limited to temperature sensors, pressure sensors, acoustic sensors, vision sensors, image sensors, and so on.
Table 1 shows an exemplary training dataset having m number of training service instance datasets and each exemplary service instance records comprising sensor values obtained from n number of sensors present in the machine. As shown in Table 1, each row is the service instance record having sensor values (S1, S2, …, Sn) and the annotated RUL value for the single time-step and it represents a single data point.
Instance_ID Time-step Sensor 1 (S1) Sensor 2 (S2) ........ Sensor n (Sn) RUL
1
1
....
1
2
2
....
2
3
...
m
Table 1
At step 204 of the method 200, the one or more hardware processors 104 of the system 100 are configured to extract a training service instance sub-dataset for each of the one or more training service instance datasets, based on a predefined selection percentage value. Each training service instance sub-dataset comprises two or more service instance records out of the plurality of service instance records present in the corresponding training service instance dataset. The predefined selection percentage value is used to extract the more service instance records that are present at the end, of the corresponding training service instance dataset i.e., the service instance records that are just close to the maintenance, repair, non-working of the machine and typically having unhealthy status and the remaining service instance records are considered to be healthy. Hence, the predefined selection percentage value is used to divide each training service instance dataset into healthy records (service instance records) and unhealthy records (service instance records), where the unhealthy records (service instance records), are captured to the corresponding training service instance sub-dataset.
For example, if the length of the training service instance dataset is 200 (having 200 service instance records) and the predefined selection percentage value (v) is 30, then the last 30% service instance records (data points or time-steps) i.e., the last 60 service instance records are extracted to the training service instance sub-dataset. Let the length of the i^th training instance after v% data sub-selection (from the last) is l(i), then the service instance records present in the corresponding training service instance sub-dataset can be represented as a sequence of tuples as follows:
= {(x_1^i,r_1^i),(x_2^i,r_2^i),?,(x_t^i,r_t^i),?,(x_(l(i))^i,r_(l(i))^i)}
where, i=1,2,3,…,k
t=1,2,3,…,l(i)
k is the length of the training service instance sub-dataset
x_t^i= feature vector at the time-step ‘t’ for the i^th service instance record
and r_t^i= annotated RUL value at the time-step ‘t’ for the i^th service instance record
Hence, the service instance records that are extracted to the training service instance sub-dataset are considered as unhealthy status records or near unhealthy status records and the service instance records that are not extracted to the training service instance sub-dataset are considered as healthy status records or near healthy status records.
At step 206 of the method 200, the one or more hardware processors 104 of the system 100 are configured to form one or more service instance record subsequences, for each training service instance sub-dataset, based on a predefined subsequence window size. Each service instance record subsequence comprises two or more service instance records out of the two or more service instance records present in the corresponding training service instance sub-dataset. The predefined subsequence window size is used to define the number of the two or more service instance records present in each service instance record subsequence and hence the number of the two or more service instance records present in each service instance record subsequence is equal to the predefined subsequence window size. In an embodiment, the predefined subsequence window size is less than or equal to the length of the training service instance sub-dataset.
For example, if the length of the training service instance sub-dataset includes 60 service records and if the predefined subsequence window size 50 then, the first service instance subsequence includes the service instance records from 1 to 50 (the rows or time-steps, or data points), the second service instance subsequence includes the service instance records from 2 to 51, and so on the last service instance subsequence includes the service instance records from 11 to 60. Hence, 11 such service instance subsequences are formed from the 60 service records having the predefined subsequence window size as 50.
If w is the predefined subsequence window size, then the length of the formed subsequences (number of the formed subsequences), k(i)= l(i)-w+1; i=1,2,…,k)
The service instance record subsequences can be represented as {S_q^i}
where S_q^i=q^th subsequence generated from the i^th training service instance dataset;
q=1,2,…,k(i); and
|S_q^i |=w
At step 208 of the method 200, the one or more hardware processors 104 of the system 100 are configured to form one or more time-series snippets for each service instance record subsequence of the one or more service instance record subsequences formed at step 206 of the method 200, to obtain a plurality of time-series snippets from the one or more training service instance datasets, using a predefined time-series snippet length. Each time-series snippet comprises the number of the service instance records equal to the predefined time-series snippet length. The predefined time-series snippet length is used further to define the number of the service instance records present in each time-series snippet of time-series snippets for each service instance record subsequence. Hence the number of the service instance records present in each time-series snippet is equal to the predefined time-series snippet length. In an embodiment, the predefined time-series snippet length is less than or equal to the length of the service instance record subsequence.
For example, if the service instance subsequence includes the service instance records from 2 to 51 and if the predefined time-series snippet length is 25, then, the first time-series snippet includes the service instance records from 2 to 26 (the rows or time-steps, or data points), the second time-series snippet includes the service instance records from 3 to 27, and so on the last time-series snippet includes the service instance records from 27 to 51. Hence, 26 such time-series snippets are formed from the service instance subsequence includes the service instance records from 2 to 51 and having the predefined time-series snippet length as 25. Likewise, the plurality of time-series snippets is formed from the one or more training service instance datasets.
At step 210 of the method 200, the one or more hardware processors 104 of the system 100 are configured to train a supervised contrastive regressive encoder, using the plurality of time-series snippets formed at step 208 of the method 200, to obtain a trained contrastive encoder. The supervised contrastive regressive encoder includes typically 2 replicas of contrastive encoders and are trained using the plurality of time-series snippets as the training data. FIG. 2D illustrates an exemplary flow diagram for training the supervised contrastive regressive encoder, in accordance with some embodiments of the present disclosure. As shown in FIG. 2D, the training process of the supervised contrastive regressive encoder, using the plurality of time-series snippets, is explained in detail through steps 210a through 210b4.
At step 210a, one or more training batches, are formed from the plurality of time-series snippets, based on a predefined training batch size. Each training batch includes a number of the time-series snippets equal to the predefined training batch size. In an embodiment, the predefined training batch size is in multiple of 2 and is defined based on the network configuration of the supervised contrastive regressive encoder. In an embodiment, the exemplary value of the predefined training batch size is 16. Further, the predefined training batch size is decided based on the number of the plurality of time-series snippets and the predefined training batch size is less than or equal to the number of the plurality of time-series snippets. It is to be understood that the last training batch may not have the number of the time-series snippets equal to the predefined training batch size.
At step 210b, the supervised contrastive regressive encoder is trained with the time-series snippets present in each training batch, at a time, until the one or more training batches are completed to obtain the trained contrastive encoder. The training process of the supervised contrastive regressive encoder with the time-series snippets present in each training batch is further explained through steps 210b1 through 210b4.
At step 210b1, each time-series snippet present in the training batch, is passed to the supervised contrastive regressive encoder, to obtain the one or more encoded features of the corresponding time-series snippet.
At step 210b2, a loss function value for each time-series snippet, is calculated based on a similarity of the corresponding time-series snippet with each other time-series snippet present in the training batch, using the corresponding one or more encoded features. In an embodiment, the similarity of two time-series snippets is calculated based on a statistical RUL value associated with each time-series snippet. In an embodiment, the statistical RUL value associated with each time-series snippet is obtained by computing a statistical characteristic among the RUL values of the service instance records present in the corresponding time-series snippet. In an embodiment, the statistical characteristic is one of mean, and a median.
Firstly, a set of time-series snippet pairs is generated from the time-series snippets present in the training batch. Let, that set be denoted as A_q^i and |A_q^i |=r i.e., the set contains r pairs of time-series snippets. Then, that set can be represented as
A_q^i= {p_(q,1)^i,p_(q,2)^i,?,p_(q,j)^i,?,p_(q,r)^i} = {p_(q,j)^i} ............................................ (1) where p_(q,j)^i = j^thtime-series snippet pair for the q^th service instance record subsequence of i^thtraining service instance dataset.
i = 1, 2, ..., m;
q = 1, 2, ..., k(i); and
j = 1, 2, ..., r;
Now, the pair p_(q,j)^i can be further represented in terms of the individual time-series snippet as follows:
p_(q,j)^i=(U_(q,2j-1)^i,U_(q,2j)^i) .............................................................................. (2)
where U_(q,2j-1)^i and U_(q,2j)^i are the individual time-series snippets with the ranges of i,q, and j are same as stated in the equation (1)
and the length of U_(q,2j-1)^i = the length of U_(q,2j)^i = s (say)
Then, the similarity of the time-series snippets present in each time-series snippet pair is calculated based on Intra-pair RUL criterion. The time-series snippets present in each time-series snippet pair, are said to be similar, when such time-series snippets are closer to each-other in terms of the Remaining Useful Life (RUL) criterion. Such pairs are referred to as positive pairs. Similarly, the time-series snippets present in each time-series snippet pair, are said to be not similar, when such time-series snippets are not closer to each-other in terms of the Remaining Useful Life (RUL) criterion. Such pairs are referred as negative pairs.
The positive pair of U_(q,2j-1)^i and U_(q,2j)^i are formed by the following steps:
First U_(q,2j-1)^i is formed by selecting a time-series snippet of length = s from the subsequence S_q^i .
Now, consider the annotated RUL values corresponding to U_(q,2j-1)^i. Find the maximum (r_max), minimum (r_min) and median (r_med) of them.
Next U_(q,2j)^i is formed by selecting another time-series snippet of the same length s from the same subsequence S_q^i such that its each RUL value (say, r') and the median of the RUL values (r_med^') satisfy the following conditions:
r'_med?r_med
r_min-d=r^'=r_max+d
where d=ß"%" of r_med =0.01ß*r_med
ß can be treated as a hyper-parameter. e.g., ß=5
Hence, the similarity (Sim () function) between two time-series snippets (TS) with respect to their RUL is determined as follows:
Similarity of two TS snippet would be 0 if the absolute value of their difference of RUL is greater than equal to the RUL of the reference TS snippet.
Else the similarity is defined as the normalized absolute difference of the RUL value of reference TS snippet and the other TS snippet under consideration.
For example, let’s consider 3 pairs of time-series tuples defined with the corresponding RUL values as {(TS1, 20), (TS2, 22)}; {(TS3, 5), (TS4, 6)}; {(TS5, 18), (TS6, 19)}. Now, the similarity Sim () is calculated as:
Sim (TS1, TS2) = 0.91;
Sim (TS1, TS5) = 0.88;
Sim (TS1, TS6) = 0.95
Sim (TS1, TS3) = Sim (TS3, TS4) = 0
So, the similarity (i.e., the Sim () function value) of TS1 will be high with respect to. TS2, TS5 and TS6 and on the other hand TS1 will be less similar with respect to TS3 and TS4.
During the training of the supervised contrastive regressive encoder, two time-series snippets are passed to the contrastive encoders and the similarity between them is calculated based on the corresponding RUL values, using the encoding features. For example, consider the time-series snippets with the RUL values as: {(TS1, 20), (TS2, 22), (TS3, 5), (TS4, 6), (TS5, 18), (TS6, 19)}. Then firstly, the pairs {(TS1, 20), (TS2, 22)}, {(TS1, 20), (TS3, 5)}, {(TS1, 20), (TS4, 6)}, ({(TS1, 20), (TS5, 18)}, and {(TS1, 20), (TS6, 19)} are passed to the supervised contrastive regressive encoder to learn the respective similarity. Next, {(TS2, 22), (TS3, 5)}, {(TS2, 22), (TS4, 6)}, {{(TS2, 22), (TS5, 18)}, and {(TS2, 22), (TS6, 19)} are passed to the supervised contrastive regressive encoder to learn the respective similarity, and so on, until all such unique time-series pairs are compared for their similarity.
The main concept of the training of the supervised contrastive regressive encoder especially in the prognostic domain is that the time-series snippets in the nearby temporal space shows similarity with respect to the Remaining Useful Life (RUL) criterion whereas the reverse is true when the time-series snippets are comparatively larger distance apart in the temporal space from the each-other. Like this, the loss function value for each time-series snippet, is calculated based on the similarity of the corresponding time-series snippet with each other time-series snippet present in the training batch.
At step the loss function value of the training batch, is calculated by adding the loss function value of each time-series snippet present in the training batch. The loss function value of each training batch is to be minimized during the training process. The loss function value of the training batch or the loss function (L_SRCL) of the supervised contrastive regressive encoder is defined as in equation 3:
L_SRCL=?_(i?I)¦L_(?SRCL?_i ) =?_(i?I)¦(-1)/(?_¦(j?I@j?i)¦?"Sim" (j,i)?) ?_¦(j?N ~(i)@j?i)¦?log (exp(z_i.z_j/t))/(?_¦(k?I@k?i)¦?exp(z_i.z_k/t)?)? ……….(3)
Where i,j are the time-series snippets and the z_i,z_j are the respective one or more encoded features. The z_k represents the one or more encoded features of all the time-series snippets present in the training batch.
At step 210b4, network parameters of the supervised contrastive regressive encoder, are updated based on the loss function value of the training batch obtained at step 210b3 and a predetermined loss function threshold value. If the loss function value of the training batch obtained at step 210b3 is less than or equal to the predetermined loss function threshold value, then the network parameters of the supervised contrastive regressive encoder are updated through the back propagation. If the loss function value of the training batch obtained at step 210b3 is greater than the predetermined loss function threshold value, then the network parameters of the supervised contrastive regressive encoder remain the same for the next training batch. Similarly, the supervised contrastive regressive encoder is trained until the one or more training batches are completed to obtain the trained contrastive encoder.
At step 210 of the method 200, the one or more hardware processors 104 of the system 100 are configured to assign a classification label for each service instance record of the plurality of service instance records present in each of the one or more training service instance datasets, received at step 202 of the method 200. The classification label one of (i) a healthy status and (ii) an unhealthy status. The classification label for each service instance record is assigned using the predefined selection percentage value as defined at step 204 of the method 200.
As explained at step 204 of the method 200, the predefined selection percentage value is used to extract the more service instance records that are present at the end, of the corresponding training service instance dataset i.e., the service instance records that are just close to the maintenance, repair, non-working of the machine and typically having unhealthy status and the remaining service instance records are considered to be healthy. Hence, the predefined selection percentage value is used to divide each training service instance dataset into healthy records (service instance records) and unhealthy records (service instance records), where the unhealthy records (service instance records), are captured to the corresponding training service instance sub-dataset.
Hence, the service instance records that are extracted to the training service instance sub-dataset at step 204 of the method are assigned with the classification label as ‘unhealthy status’ and the service instance records that are not extracted to the training service instance sub-dataset are assigned with the classification label as ‘healthy status’.
At step 214 of the method 200, the one or more hardware processors 104 of the system 100 are configured to obtain the one or more encoded features for each service instance record of the plurality of service instance records present in each of the one or more training service instance datasets, received at step 202 of the method 200. The one or more encoded features for each service instance record are obtained by passing the corresponding one or more sensor values to the trained contrastive encoder obtained at step 210 of the method 200.
At step 216 of the method 200, the one or more hardware processors 104 of the system 100 are configured to train a classifier, with the plurality of service instance records present in each of the one or more training service instance datasets received at step 202 of the method 200, to obtain a trained classifier. The classifier is trained with the corresponding one or more encoded features obtained at step 214 of the method 200 of each service instance records as input variables and the corresponding classification label assigned at step 214 of the method 200 to each service instance record as output variable. The training process of the classifier is generic and employ one of the conventional techniques for the training. In an embodiment, the classifier is one of classification network from the list including but are not limited to random forest classifier and support vector machine (SVM) classifier.
At step 218 of the method 200, the one or more hardware processors 104 of the system 100 are configured to train a regressor, with the two or more service instance records present in each training service instance sub-dataset of the one or more training service instance datasets obtained at step 204 of the method 200, to obtain a trained regressor. The regressor is trained with the corresponding one or more encoded features obtained at step 214 of the method 200 of each service instance records as input variables and the corresponding RUL value received at step 202 of the method 200 to each service instance record as output variable. The training process of the regressor is generic and employs one of the conventional techniques for the training. In an embodiment, the regressor is one of regressor network from the list including but are not limited to random forest regressor and support vector machine (SVM) regressor.
FIG. 3A illustrates an exemplary flow diagram for obtaining a set of models to estimate remaining useful life of a machine, in accordance with some embodiments of the present disclosure. The set of models includes the trained contrastive encoder obtained at step 210 of the method 200, the trained classifier obtained at step 216 of the method 200, and the trained regressor obtained at step 218 of the method 200. Obtaining the trained contrastive encoder, the trained classifier, and the trained regressor, is one time activity and the obtained trained contrastive encoder, the trained classifier and the trained regressor are then used to estimate the prognosis and the remaining useful time (RUL) of the machine in real time using the corresponding sensor data.
At step 220 of the method 200, the one or more hardware processors 104 of the system 100 are configured to receive one or more test service instance records of the machine whose health to be monitored. Each test service instance record includes one or more sensor test values and are obtained using the one or more sensors present in such machine.
At step 222 of the method 200, the one or more hardware processors 104 of the system 100 are configured to obtain the one or more encoded features, for each test service instance record of the one or more test service instance records received at step 220 of the method 200. The one or more encoded features for each test service instance record are obtained by passing the corresponding one or more sensor test values to the trained contrastive encoder obtained at step 210 of the method 200.
At step 224 of the method 200, the one or more hardware processors 104 of the system 100 are configured to pass the one or more encoded features obtained for each test service instance record of the one or more test service instance records obtained at step 222 of the method 200, to the trained classifier, to predict one of: (i) the healthy status and (ii) the unhealthy status, of the machine. If the health of the machine predicted as the healthy status, then no need to take any precautionary actions. Else, the remaining useful life (RUL) of the machine to be determined for taking the timely precautionary actions.
At step 226 of the method 200, the one or more hardware processors 104 of the system 100 are configured to pass the one or more encoded features obtained for each test service instance record of the one or more test service instance records obtained at step 222 of the method 200, to the trained regressor, to predict the RUL value of the machine. The remaining useful life (RUL) of the machine is determined for taking the timely precautionary actions if the machine is predicted as the unhealthy status during the classification at step 224 of the method 200.
FIG. 3B illustrates an exemplary flow diagram for estimating remaining useful life of the machine using the test data through the set of models, in accordance with some embodiments of the present disclosure. As shown in FIG. 3B, using the set of models, the health status and the RUL value of the machine can be estimated to take appropriate measures.
The embodiments of present disclosure herein address unresolved problem of estimating the remaining useful life (RUL) of a machine, using the temporal sensor data. According to the present disclosure, the temporal sensor data is pre-processed in such a way that the supervised contrastive encoder is able to increase the distance among the candidates belonging to the different classes in the embedded space to address supervised multiclass classification problem. The supervised multiclass classification classifies the machine with a healthy status or unhealthy status using the embedded space obtained by the contrastive learning. Further, the regression model is employed to estimate the RUL of the machine using the embedded space obtained by the contrastive learning if the machine is said to be with the unhealthy status. Hence the estimation is the RUL of the machine according to the present disclosure is accurate and efficient.
Example scenario:
The present disclosure has been tested with 2 different client specific datasets (training and testing service instance datasets) which are vehicle telemetry data (i.e., multi-sensor prognostic datasets), referred as datasets as dataset A and dataset B. The dataset A is comparatively smaller than the dataset B. The dataset A has the training service instance datasets of 12 associated with each of 132 different vehicle engines, i.e., 1584 training service instance datasets and has one testing service instance datasets associated with each of 92 different vehicle engines, i.e., 92 testing service instance datasets. Table 2 shows the experimental results in terms of a root mean square error (RMSE) of the trained regressor of the present disclosure and the conventional DL approach such as long short-term memory (LSTM) based approach for the training service instance datasets and the one testing service instance datasets of the dataset A.
Present disclosure Conventional DL based Approach
Train RMSE 4.8 8.6
Test RMSE 5.1 9.1
Table 2
The dataset A has the training service instance datasets of 11 associated with each of 328 different vehicle engines, i.e., 3608 training service instance datasets and has one testing service instance datasets associated with each of 152 different vehicle engines, i.e., 152 testing service instance datasets. Table 3 shows the experimental results in terms of the root mean square error (RMSE) of the trained regressor of the present disclosure and the conventional DL approach for the training service instance datasets and the one testing service instance datasets of the dataset B.
Present disclosure Conventional DL based Approach
Train RMSE 3.8 6.9
Test RMSE 4.1 7.2
Table 3
From Table 2 and Table 3, it is evident that the trained regressor of the present disclosure outperforms compared to the conventional DL based approach. Hence the present disclosure is efficient and accurate for estimating the remaining useful life (RUL) of the machine.
The written description describes the subject matter herein to enable any person skilled in the art to make and use the embodiments. The scope of the subject matter embodiments is defined by the claims and may include other modifications that occur to those skilled in the art. Such other modifications are intended to be within the scope of the claims if they have similar elements that do not differ from the literal language of the claims or if they include equivalent elements with insubstantial differences from the literal language of the claims.
It is to be understood that the scope of the protection is extended to such a program and in addition to a computer-readable means having a message therein; such computer-readable storage means contain program-code means for implementation of one or more steps of the method, when the program runs on a server or mobile device or any suitable programmable device. The hardware device can be any kind of device which can be programmed including e.g., any kind of computer like a server or a personal computer, or the like, or any combination thereof. The device may also include means which could be e.g., hardware means like e.g., an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination of hardware and software means, e.g., an ASIC and an FPGA, or at least one microprocessor and at least one memory with software processing components located therein. Thus, the means can include both hardware means and software means. The method embodiments described herein could be implemented in hardware and software. The device may also include software means. Alternatively, the embodiments may be implemented on different hardware devices, e.g., using a plurality of CPUs.
The embodiments herein can comprise hardware and software elements. The embodiments that are implemented in software include but are not limited to, firmware, resident software, microcode, etc. The functions performed by various components described herein may be implemented in other components or combinations of other components. For the purposes of this description, a computer-usable or computer readable medium can be any apparatus that can comprise, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
The illustrated steps are set out to explain the exemplary embodiments shown, and it should be anticipated that ongoing technological development will change the manner in which particular functions are performed. These examples are presented herein for purposes of illustration, and not limitation. Further, the boundaries of the functional building blocks have been arbitrarily defined herein for the convenience of the description. Alternative boundaries can be defined so long as the specified functions and relationships thereof are appropriately performed. Alternatives (including equivalents, extensions, variations, deviations, etc., of those described herein) will be apparent to persons skilled in the relevant art(s) based on the teachings contained herein. Such alternatives fall within the scope of the disclosed embodiments. Also, the words “comprising,” “having,” “containing,” and “including,” and other similar forms are intended to be equivalent in meaning and be open ended in that an item or items following any one of these words is not meant to be an exhaustive listing of such item or items, or meant to be limited to only the listed item or items. It must also be noted that as used herein and in the appended claims, the singular forms “a,” “an,” and “the” include plural references unless the context clearly dictates otherwise.
Furthermore, one or more computer-readable storage media may be utilized in implementing embodiments consistent with the present disclosure. A computer-readable storage medium refers to any type of physical memory on which information or data readable by a processor may be stored. Thus, a computer-readable storage medium may store instructions for execution by one or more processors, including instructions for causing the processor(s) to perform steps or stages consistent with the embodiments described herein. The term “computer-readable medium” should be understood to include tangible items and exclude carrier waves and transient signals, i.e., be non-transitory. Examples include random access memory (RAM), read-only memory (ROM), volatile memory, nonvolatile memory, hard drives, CD ROMs, DVDs, flash drives, disks, and any other known physical storage media.
It is intended that the disclosure and examples be considered as exemplary only, with a true scope of disclosed embodiments being indicated by the following claims.
, Claims:We Claim:
1. A processor-implemented method (200), comprising the steps of:
receiving, via one or more hardware processors, one or more training service instance datasets associated with a machine, wherein each training service instance dataset comprises a plurality of service instance records and wherein each service instance record is a time-series data and comprises one or more sensor values and a corresponding RUL value (202);
extracting, via the one or more hardware processors, a training service instance sub-dataset for each of the one or more training service instance datasets, based on a predefined selection percentage value, wherein each training service instance sub-dataset comprises two or more service instance records out of the plurality of service instance records present in the corresponding training service instance dataset (204);
forming, via the one or more hardware processors, one or more service instance record subsequences, for each training service instance sub-dataset, based on a predefined subsequence window size, wherein each service instance record subsequence comprises the two or more service instance records and a number of the two or more service instance records present in each service instance record subsequence is equal to the predefined subsequence window size (206);
forming, via the one or more hardware processors, one or more time-series snippets for each service instance record subsequence of the one or more service instance record subsequences, to obtain a plurality of time-series snippets from the one or more training service instance datasets, using a predefined time-series snippet length, wherein each time-series snippet comprises the number of the service instance records equal to the predefined time-series snippet length (208);
training, via the one or more hardware processors, a supervised contrastive regressive encoder, using the plurality of time-series snippets, to obtain a trained contrastive encoder (210);
assigning, via the one or more hardware processors, a classification label from one of (i) a healthy status and (ii) an unhealthy status, for each service instance record of the plurality of service instance records present in each of the one or more training service instance datasets, using the predefined selection percentage value (212);
obtaining, via the one or more hardware processors, one or more one or more encoded features, for each service instance record of the plurality of service instance records present in each of the one or more training service instance datasets, by passing the corresponding one or more sensor values to the trained contrastive encoder (214);
training, via the one or more hardware processors, a classifier, with the plurality of service instance records present in each of the one or more training service instance datasets, using the one or more encoded features and the corresponding classification label of each service instance record, to obtain a trained classifier (216); and
training, via the one or more hardware processors, a regressor, with the two or more service instance records present in each training service instance sub-dataset of the one or more training service instance datasets, using the corresponding one or more encoded features and the corresponding RUL value of each service instance, to obtain a trained regressor (218).

2. The processor-implemented method of claim 1, further comprising:
receiving, via the one or more hardware processors, one or more test service instance records of a machine whose health to be monitored, wherein each test service instance record comprises one or more sensor test values (220);
obtaining, via the one or more hardware processors, the one or more one or more encoded features, for each test service instance record of the one or more test service instance records, by passing the corresponding one or more sensor test values to the trained contrastive encoder (222);
passing, via the one or more hardware processors, the one or more one or more encoded features obtained for each test service instance record of the one or more test service instance records, to the trained classifier, to predict one of: (i) the healthy status and (ii) the unhealthy status, of the machine (224); and
passing, via the one or more hardware processors, the one or more encoded features obtained for each test service instance record of the one or more test service instance records, to the trained regressor, to predict the RUL value of the machine, if the machine is predicted as the unhealthy status during the classification (226).

3. The processor-implemented method of claim 1, wherein the one or more sensor values are obtained from one or more sensors present in the machine.

4. The processor-implemented method of claim 1, wherein each training service instance dataset associated with the machine, is indicative of servicing data of the machine, just before an abnormal condition is encountered.

5. The processor-implemented method of claim 1, wherein training the supervised contrastive regressive encoder, using the plurality of time-series snippets, to obtain the trained contrastive encoder, comprises:
forming one or more training batches, from the plurality of time-series snippets, based on a predefined training batch size, wherein each training batch comprises a number of the time-series snippets equal to the predefined training batch size (210a); and
training the supervised contrastive regressive encoder with the time-series snippets present in each training batch, at a time, until the one or more training batches are completed to obtain the trained contrastive encoder (210b), wherein training the supervised contrastive regressive encoder with the time-series snippets present in each training batch comprises:
passing each time-series snippet present in the training batch, to the supervised contrastive regressive encoder, to obtain the one or more encoded features of the corresponding time-series snippet (210b1);
calculating a loss function value for each time-series snippet, based on a similarity of the corresponding time-series snippet with each other time-series snippet present in the training batch, using the corresponding one or more encoded features (210b2);
calculating a loss function value of the training batch, by adding the loss function value of each time-series snippet present in the training batch (210b3); and
updating network parameters of the supervised contrastive regressive encoder, based on the loss function value of the training batch and a predetermined loss function threshold value (210b4).

6. The processor-implemented method of claim 5, wherein similarity of two time-series snippets is calculated based on a statistical RUL value associated with each time-series snippet, and wherein the statistical RUL value associated with each time-series snippet is obtained by computing a statistical characteristic among the RUL values of the service instance records present in the corresponding time-series snippet.

7. A system (100) comprising:
a memory (102) storing instructions;
one or more input/output (I/O) interfaces (106); and
one or more hardware processors (104) coupled to the memory (102) via the one or more I/O interfaces (106), wherein the one or more hardware processors (104) are configured by the instructions to:
receive one or more training service instance datasets associated with a machine, wherein each training service instance dataset comprises a plurality of service instance records and wherein each service instance record is a time-series data and comprises one or more sensor values and a corresponding RUL value;
extract a training service instance sub-dataset for each of the one or more training service instance datasets, based on a predefined selection percentage value, wherein each training service instance sub-dataset comprises two or more service instance records out of the plurality of service instance records present in the corresponding training service instance dataset;
form a one or more service instance record subsequences, for each training service instance sub-dataset, based on a predefined subsequence window size, wherein each service instance record subsequence comprises the two or more service instance records and a number of the two or more service instance records present in each service instance record subsequence is equal to the predefined subsequence window size;
form one or more time-series snippets for each service instance record subsequence of the one or more service instance record subsequences, to obtain a plurality of time-series snippets from the one or more training service instance datasets, using a predefined time-series snippet length, wherein each time-series snippet comprises the number of the service instance records equal to the predefined time-series snippet length;
train a supervised contrastive regressive encoder, using the plurality of time-series snippets, to obtain a trained contrastive encoder;
assign a classification label from one of (i) a healthy status and (ii) an unhealthy status, for each service instance record of the plurality of service instance records present in each of the one or more training service instance datasets, using the predefined selection percentage value;
obtain one or more encoded features, for each service instance record of the plurality of service instance records present in each of the one or more training service instance datasets, by passing the corresponding one or more sensor values to the trained contrastive encoder;
train a classifier, with the plurality of service instance records present in each of the one or more training service instance datasets, using the corresponding one or more encoded features and the corresponding classification label of each service instance record, to obtain a trained classifier; and
train a regressor, with the two or more service instance records present in each training service instance sub-dataset of the one or more training service instance datasets, using the corresponding one or more encoded features and the corresponding RUL value of each service instance, to obtain a trained regressor.

8. The system of claim 7, wherein the one or more hardware processors (104) are further configured to:
receive one or more test service instance records of a machine whose health to be monitored, wherein each test service instance record comprises one or more sensor test values;
obtain the one or more encoded features, for each test service instance record of the one or more test service instance records, by passing the corresponding one or more sensor test values to the trained contrastive encoder;
pass the one or more encoded features obtained for each test service instance record of the one or more test service instance records, to the trained classifier, to predict one of: (i) the healthy status and (ii) the unhealthy status, of the machine; and
pass the one or more encoded features obtained for each test service instance record of the one or more test service instance records, to the trained regressor, to predict the RUL value of the machine, if the machine is predicted as the unhealthy status during the classification.

9. The system of claim 7, wherein the one or more sensor values are obtained from one or more sensors present in the machine.

10. The system of claim 7, wherein each training service instance dataset associated with the machine, is indicative of servicing data of the machine, just before an abnormal condition is encountered.

11. The system of claim 7, wherein the one or more hardware processors (104) are configured to train the supervised contrastive regressive encoder, using the plurality of time-series snippets, to obtain the trained contrastive encoder, by:
forming one or more training batches, from the plurality of time-series snippets, based on a predefined training batch size, wherein each training batch comprises a number of the time-series snippets equal to the predefined training batch size; and
training the supervised contrastive regressive encoder with the time-series snippets present in each training batch, at a time, until the one or more training batches are completed to obtain the trained contrastive encoder, wherein training the supervised contrastive regressive encoder with the time-series snippets present in each training batch comprises:
passing each time-series snippet present in the training batch, to the supervised contrastive regressive encoder, to obtain the one or more encoded features of the corresponding time-series snippet;
calculating a loss function value for each time-series snippet, based on a similarity of the corresponding time-series snippet with each other time-series snippet present in the training batch, using the corresponding one or more encoded features;
calculating a loss function value of the training batch, by adding the loss function value of each time-series snippet present in the training batch; and
updating network parameters of the supervised contrastive regressive encoder, based on the loss function value of the training batch and a predetermined loss function threshold value.

12. The system of claim 7, wherein the one or more hardware processors (104) are configured to calculate similarity of two time-series snippets, based on a statistical RUL value associated with each time-series snippet, and wherein the statistical RUL value associated with each time-series snippet is obtained by computing a statistical characteristic among the RUL values of the service instance records present in the associated time-series snippet.

Dated this 22nd Day of December 2022

Tata Consultancy Services Limited
By their Agent & Attorney

(Adheesh Nargolkar)
of Khaitan & Co
Reg No IN-PA-1086