Description:Complete Specification:
The following specification describes and ascertains the nature of this invention and the manner in which it is to be performed

Field of the invention
[0001] The present disclosure relates to the field of data analytics and predictive diagnostics. In particular, the disclosure describes a system and method to monitor tool wear.

Background of the invention
[0002] In complex industrial systems such as automobile or aerospace systems, that contain multitude of sensors, root cause analysis of a fault is tough. This is because the identification of fault signatures and which sensors contributed to the fault are also difficult to establish. Moreover, different sensors capture different types of data (image, sound, ultrasound, pressure, temperature, velocity etc.). Currently, methods proposed in literature analyze these signals independently and identify abnormalities in their signatures independent of other signals, thereby causing high false positives. There is a need for an adaptive ensemble technique to provide a flexible and intelligent framework which not only combines features across different types of data (and sensors) but also allows the intelligent system to prioritize different inputs for identifying different failure modes thereby making it robust to capture most faults with minimal false positives.

[0003] Patent Application US2019227528 AA titled “Power tool including a machine learning block” proposes a power tool includes a housing and a sensor, a machine learning controller, a motor, and an electronic controller supported by the housing. The sensor is configured to generate sensor data indicative of an operational parameter of the power tool. The machine learning controller includes a first processor and a first memory and is coupled to the sensor. The machine learning controller further includes a machine learning control program configured to receive the sensor data, process the sensor data using the machine learning control program, and generate an output based on the sensor data using the machine learning control program. The electronic controller includes a second processor and a second memory and is coupled to the motor and to the machine learning controller. The electronic controller is configured to receive the output from the machine learning controller and control the motor based on the output.

Brief description of the accompanying drawings
[0004] An embodiment of the invention is described with reference to the following accompanying drawings:
[0005] Figure 1 depicts a system (100) to monitor tool (10) wear;
[0006] Figure 2 illustrates method steps (200) to monitor tool (10) wear using an AI based system (100).

Detailed description of the drawings

[0007] Figure 1 depicts a system (100) to monitor tool (10) wear. A tool (10) with reference to this disclosure refers to a hardware equipment, instrument or a component used in any industrial system (100). Figure is an exemplary embodiment that illustrates the system (100) in the context of a power tool (10) operating on a workpiece (14) placed on a machining table (11).

[0008] The system (100) to monitor tool (10) wear comprises a plurality of sensors (12) adapted to monitor a plurality of tool (10) operating parameters, a processing unit (16) in communication with the plurality of sensors (12) and at least an output interface (18).

[0009] The plurality of sensors (12) may be internal sensors (12), which are arranged in the housing of the power tool (10), or external sensors (12), which are arranged on or outside the housing of the power tool (10), for example on the machining table. The different types of sensors (12) include but are not limited to a motion sensor, in particular an accelerometer or a gyro sensor, a temperature sensor, such as for example an NTC or a PTC, a current sensor, a rotational speed sensor, a sound sensor such as a microphone, a pressure sensor, a capacitive or resistive force sensor, an optical Sensor, for example a camera and the like. The different types of sensors (12) measure various tool (10) operating parameters. This raw data pertaining to the tool (10) operating parameters are then communicated to the processing unit (16).

[0010] The processing unit (16) can either be a logic circuitry or a software residing in a cloud that responds to and processes logical instructions to get a meaningful result. A hardware processing unit (16) may be implemented as any or a combination of: one or more microchips or integrated circuits interconnected using a parent board, hardwired logic, software stored by a memory device and executed by a microprocessor, firmware, an application specific integrated circuit (ASIC), and/or a field programmable gate array (FPGA). In an exemplary embodiment of the present disclosure the processing unit (16) executes a plurality of AI models (161).

[0011] An AI model with reference to this disclosure can be defined as reference or an inference set of data, which is use different forms of correlation matrices. Using these models and the data from these models, correlations can be established between different types of data to arrive at some logical understanding of the data. A person skilled in the art would be aware of the different types of AI models (161) such as linear regression, naïve bayes classifier, support vector machine, neural networks and the like. The AI model may be implemented as a set of software instructions, combination of software and hardware or any combination of the same. For example, neural network chips are specialized silicon chips, which incorporate AI technology and are used for machine learning.

[0012] The processing unit (16) is characterized by it’s functionality. The processing unit (16) configured to: acquire and preprocess data of the plurality of tool (10) operating parameters received from the plurality of sensors (12); extract a wear prediction factor (P1,P2,…..Pj) from the pre-processed data, wherein each AI model is configured to extract a wear prediction factor (P1,P2,…..Pj) from the pre-processed data of one type of sensor(s); assign a weight (g1,g2,…..gj) to each of the wear prediction factors (P1,P2,…..Pj);record assigned weights ((g1,g2,…..gj) and the corresponding wear prediction factors (P1,P2,…..Pj) over a period of time; calculate a tool (10) wear by aggregating the recorded values; inform at least one user of the tool (10) wear via the output interface (18).

[0013] The processing unit (16) is trained on Mixture of Experts (MoE) techniques, whereby the each of the AI model amongst the plurality of AI models (161) learns to extract the wear prediction factor based on features specific to a particular type of sensor data. The processing unit (16) based on the Mixture of Experts (MoE) techniques, learns to assign weight for a particular type of sensor to the corresponding AI model.

[0014] The functionality of the processing unit (16) further explained by means of the method steps (200) in accordance with figure 2. The processing unit (16) is in communication with the output interface (18). The output interface (18) is any audio/visual device that can be remotely located such as a tablet or mobile phone and informs a user of the tool (10) wear.

[0015] It should be understood at the outset that, although exemplary embodiments are illustrated in the figures and described below, the present disclosure should in no way be limited to the exemplary implementations and method (200) illustrated in the drawings and described below.

[0016] Figure 2 illustrates method steps (200) to monitor tool (10) wear using an AI based system (100). The system (100) and its components have been explained in accordance with figure 1. For the purpose of clarity, it is reiterated that the system (100) comprises a plurality of sensors (12) adapted to monitor a plurality of tool (10) operating parameters, a processing unit (16) comprising a plurality of AI models (161), said processing unit (16) in communication with the plurality of sensors (12).

[0017] Method step 201 comprises receiving data on a plurality of tool (10) operating parameters from the plurality of sensors (12). Method step 202 comprises preprocessing data received from the plurality of sensors (12) by the processing unit (16). Pre-processing data involves denoising operations on the signals received from the plurality of sensors (12) to extracts features that can be fed to the AI models (161) in the processing unit (16). For example, acoustic data received from microphone undergoes preprocessing like band pass filtering followed by extraction of Log mel spectrogram features. Likewise, vibration & cutting forces data received from the accelerometer undergoes preprocessing like wavelet packet decomposition followed by extraction of Log mel spectrogram features. Similarly, image data undergoes data augmentation to extract features that are fed to the AI models (161).

[0018] Method step 203 comprises extracting a wear prediction factor (P1,P2,…..Pj) from the pre-processed data, wherein each AI model configured to extract a wear prediction factor (P1,P2,…..Pj) from the pre-processed data of one type of sensor(s). The processing unit (16) is trained on Mixture of Experts (MoE) techniques, whereby the each of the AI model amongst the plurality of AI models (161) learns to extract the wear prediction factor based on features specific to a particular type of sensor data.

[0019] Method step 204 comprises assigning a weight (g1,g2,…..gj) to each of the wear prediction factors (P1,P2,…..Pj) by means of the processing unit (16). The processing unit (16) based on the Mixture of Experts (MoE) techniques, learns assign weight for a particular type of sensor to the corresponding AI model.

[0020] Mixture of Experts (MoE), learns from multiple input signals concurrently in a supervised manner during training. An expert model (AI model specific for a particular type of sensor) is trained on different sets of input signals. The framework assigns different weights or probabilities to the outputs of each expert model as a function of the input data (adaptive) and the final prediction is a weighted function of the outputs of all the expert models. The MoE framework is based on the “divide and conquer” principle which sub-divides the input space into distinct regions and trains an expert model on each input subset.

[0021] The framework learns from the training data which signals are more useful in which regions of the input space and also accounts for the mutual influence of the multiple signals. The proposed framework extends the MoE to include an extra layer of meta learner, thereby producing a Hierarchical Mixture of Experts (HME). Similar to adding additional hidden layers in a neural network to make it deep and help it learn more features, additional meta layers in HME help the framework to better adapt to different combinations of the input signals.

[0022] Method step 205 comprises recording the assigned weights ((g1,g2,…..gj) and the corresponding wear prediction factors (P1,P2,…..Pj) over a period of time by means of the processor. Method step 206 comprises calculating a tool (10) wear by aggregating the recorded values. The output of each expert model is treated as a probability score (Pij) and the weights learnt with a final Softmax function to combine the probabilities of each expert. The mathematical details of HME is as below:

[0023] A person skilled in the art will appreciate that while these method steps describes only a series of steps to accomplish the objectives, these methodologies may be implemented with modification and adaptation to the system (100) and its components. The method is part/machine agnostic – can be scaled to any number of machines .

[0024] The proposed system (100) and method to monitor tool (10) wear proposes a novel robust method for condition monitoring of complex system (100)s in key business application areas such as automotive, aerospace, industrial applications. The proposed system (100) and method adaptively learns which method is suitable for which type of errors. The framework also attaches a probabilistic score for prediction from each method to deliver the final prediction thereby effectively catching the faults whilst reducing the false positive rate. The expert AI models (161) for specific types of sensor remove the necessity for relying on expert domain knowledge of each type of fault and what its causes are (or which signals can pick up these faults).

[0025] It must be understood that the embodiments explained in the above detailed description are only illustrative and do not limit the scope of this invention. Any modification and adaptation of the system (100) and method to monitor tool (10) wear are envisaged and form a part of this invention. The scope of this invention is limited only by the claims.

, Claims:We Claim:
1. A system (100) to monitor tool (10) wear, the system (100) comprising a plurality of sensors (12) adapted to monitor a plurality of tool (10) operating parameters, a processing unit (16) in communication with the plurality of sensors (12), characterized in that system (100):
the processing unit (16) configured to acquire and preprocess data of the plurality of tool (10) operating parameters received from the plurality of sensors (12), the processing unit (16) comprising a plurality of AI models (161), the processing unit (16) configured to:
extract a wear prediction factor (P1,P2,…..Pj) from the pre-processed data, wherein each AI model is configured to extract a wear prediction factor (P1,P2,…..Pj) from the pre-processed data of one type of sensor(s);
assign a weight (g1,g2,…..gj) to each of the wear prediction factors (P1,P2,…..Pj);
record assigned weights ((g1,g2,…..gj) and the corresponding wear prediction factors (P1,P2,…..Pj) over a period of time;
calculate a tool (10) wear by aggregating the recorded values;
an output interface (18) configured to inform at least one user of the tool (10) wear.

2. The system (100) to predict tool (10) wear as claimed in claim 1, wherein the processing unit (16) is trained on Mixture of Experts (MoE) techniques, whereby the each of the AI model amongst the plurality of AI models (161) learns to extract the wear prediction factor based on features specific to a particular type of sensor data.

3. The system (100) to predict tool (10) wear as claimed in claim 1, wherein the processing unit (16) based on the Mixture of Experts (MoE) techniques, learns assign weight for a particular type of sensor to the corresponding AI model.

4. A method (200) to monitor tool (10) wear using an AI based system (100), the system (100) comprising a plurality of sensors (12) adapted to monitor a plurality of tool (10) operating parameters, a processing unit (16) comprising a plurality of AI models (161), said processing unit (16) in communication with the plurality of sensors (12), the method comprising: receiving data on a plurality of tool (10) operating parameters from the plurality of sensors (12); preprocessing data received from the plurality of sensors (12) by the processing unit (16); characterized in that method:
extracting a wear prediction factor (P1,P2,…..Pj) from the pre-processed data, wherein each AI model configured to extract a wear prediction factor (P1,P2,…..Pj) from the pre-processed data of one type of sensor(s);
assigning a weight (g1,g2,…..gj) to each of the wear prediction factors (P1,P2,…..Pj) by means of the processing unit (16);
recording the assigned weights ((g1,g2,…..gj) and the corresponding wear prediction factors (P1,P2,…..Pj) over a period of time by means of the processor;
calculating a tool (10) wear by aggregating the recorded values;

5. The method (200) to monitor tool (10) wear using an AI based system (100) as claimed in claim 4, wherein the processing unit (16) is trained on Mixture of Experts (MoE) techniques, whereby the each of the AI model amongst the plurality of AI models (161) learns to extract the wear prediction factor based on features specific to a particular type of sensor data.

6. The method (200) to monitor tool (10) wear using an AI based system (100) as claimed in claim 4, wherein the processing unit (16) based on the Mixture of Experts (MoE) techniques, learns assign weight for a particular type of sensor to the corresponding AI model.