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] This invention is related to a control unit to detect at least one infield fault in an industrial machine.

Background of the invention

[0002] A major challenge in developing a fault detection algorithm for any industrial machinery is the availability of fault data for all types of faults in good enough volume. Setting up an experimental test bench for all types of machines is a costly affair and vibrations in machines depend on their size hence a model developed on an experimental test bench will fail on a bigger industrial machine. Available sensors data at maximum industrial gearbox fails to detect all possible failures when they start to appear. The gearbox is a critical component of various types of machinery. Early detection of its failures would prevent economic and human fatalities. Among current, vibration, and acoustics data, vibration data is most used in detecting faults in a gearbox majorly related to mechanical failures.

[0003] A US7945397 patent discloses A system includes a plurality of sensors configured to measure one or more characteristics of a gearbox. The system also includes a gearbox condition indicator device, which includes a plurality of sensor interfaces configured to receive input signals associated with at least one stage of the gearbox from the sensors. The gearbox condition indicator device also includes a processor configured to identify a fault in the gearbox using the input signals and an output interface configured to provide an indicator identifying the fault. The processor is configured to identify the fault by determining a family of frequencies related to at least one failure mode of the gearbox, decomposing the input signals using the family of frequencies, reconstructing a gear signal using the deconstructed input signals, and comparing the reconstructed gear signal to a baseline signal.

Brief description of the accompanying drawings
[0004] Figure 1 illustrates a control unit to detect at least one infield fault in an industrial machine in accordance with an embodiment of the invention; and
[0005] Figure 2 illustrates illustrates a flowchart of a method for detecting at least one infield fault in an industrial machine in accordance with the present invention.

Detailed description of the embodiments
[0006] Figure 1 illustrates a control unit to detect at least one infield fault in an industrial machine in accordance with an embodiment of the invention. The machine 12 comprises a sensing element 14 for detecting at least one machine parameter. The control unit 10 develops a virtual model 16 of the machine 12 and builds a virtual numerical model 18 of the machine 12 incorporating multiple real-time environments. The control unit 10 then generates data related to operating conditions of the machine 12 working by varying the at least one machine parameter and acquires different synthetic data related to the operating conditions of the machine 12 upon varying the at least one machine parameter. The control unit 10 builds and trains an intelligence model using the acquired different synthetic data in an intelligence module 20 and detects the at least one infield fault in the machine 12 in a real-time using the intelligence module 20.

[0007] Further the working of the industrial machine in different working conditions and acquisition of data is explained in detail. The control unit 10 is chosen from a group of control units comprising a microcontroller, a microprocessor, a digital circuit, an integrated chip, and the like. According to one embodiment of the invention, the control unit 10 is integrated in the machine. And according to another embodiment of the invention, the control unit 10 is a cloud repository /cloud server , which receives the data related to the machine during calibration and detects an infield fault of the machine 12 in real-time based on the received machine data during calibration.

[0008] The at least one machine parameter detected by the sensing element is any one of parameter chosen from a current, a number of vibrations in each time, a speed of said machine, a working temperature of said machine. For instance, a vibrational sensing element is used to sense the number of vibrations in a given time period. However, it is to be understood that the type of parameters are not restricted to the above disclosed one’s , but can be of any other type that is known to a person skilled in the art. The type of parameter that needs to be considered is based on the type of the industrial machine 12.

[0009] The operating condition of the machine 12 is chosen from the conditions wherein said machine is working in different speeds, temperatures, loads. For instance, the data is generated during a full load operating condition or a half load operating condition or by varying the current passing the machine 12. The generated data is acquired and is processed in the intelligence module 20. The acquisition of the different synthetic data comprises both a healthy operating condition and a faulty operating condition of the machine 12.

[0010] The intelligence module 20 is chosen from a group of intelligence modules comprising an artificial intelligence module (AI), a deep learning module (DL) and a machine learning module (ML). These modules 20 are build and trained with different types of synthetic data. I.e.., the data generated during the healthy working conditions and also when the machine working under various machine parameter levels. For instance, the data of the machine 12 is generated when the machine is working at a half load and at different current levels.

[0011] The intelligence module 20 receives data from multiple sources and at different working conditions of the machine 12. As disclosed above, the intelligence module 20 is trained with the data during the healthy working condition of the machine 12 and also during the failure conditions of the machine 12. The intelligence module 20 (i.e., having a neural network) is trained and developed with all these data.
[0012] The data can be received from the sensing element 14 (it can be one or more, that are used to detect various parameters of the machine) and the data generated during the healthy and the faulty conditions of the machine 12. All these data is verified and validated and is fed into the intelligence module 20 for real-time detection of the working state of the machine 12.

[0013] Figure 2 illustrates a flowchart of a method for detecting at least one infield fault in an industrial machine 12 according to the present invention. In step S1, at least one machine parameter of the machine is detected by a sensing element 14. In step S2, the virtual model 16 of the machine 12 is developed and a virtual numerical model 18 of the machine 12 is build incorporating multiple real-time environments by a control unit 10. In step S3, data related to operating conditions of the machine 12 working is generated by varying the at least one machine parameter. In step S4, different synthetic data related to the operating conditions of the machine 12 is acquired, upon varying at least one machine parameter. In step S5, an intelligence model is built and trained using the acquired different synthetic data in an intelligence module 20 . In step S6, the at least one infield fault in the machine 12 is detected in a real-time using the intelligence module 20.

[0014] The method is explained in detail. The control unit 10 builds or develops a virtual model 16 of the machine 12 and a virtual numerical model 18 of the machine 12 using one of the building techniques that is known in the state of the art. While building/developing the virtual numerical model 18 is build , the control unit 10 considers different real-time environments, such that, the build numerical model 18 is as close to the real -time testing equipment. The control unit 10 then verifies and validates the build numerical model 18 and checks whether the accuracy of the build numerical model 18 is within the acceptable range.

[0015] The acceptable range is predefined in a memory (not shown) of the control unit 10. Then the control unit 10 generates and acquires data related to the machine 12, wherein the machine is working is different operating conditions and also varying at least one parameter. The generated and acquired synthetic data that is received from under multiple operating conditions of the machine 12 and the data that is received from the sensing elements 14 of the machine 12 is used to build and train the intelligence module 20 . This developed intelligence module 20 is used to predict the in-field fault in the machine 20 at a very early stage.

[0016] The method disclosed above is explained by providing an example. According to one embodiment of the invention, the machine 12 is a gear box. Gear box is the critical sub asset in an industrial application. Required speed of the driven system is controlled by gearbox 12. As gearbox 12 transfer the load and speed from drive to driven system. It undergo various cyclic loadings, which can cause the failure of gear box 12 intern complete system. Know the health condition of gearbox infield is very important aspect for continuous operations.

[0017] Failures in the gearbox 12 are not instantaneous. It starts during operations and develops the symptoms over a period. If the system failed to diagnose the issue early, would lead to catastrophic failure. Available methods to find the fault in the gearbox12 purely depend on the sensor data. And most of the sensing elements 14 would fail to predict the failure pattern when the failure begins. Most of the methods are developed using data from the experimental test bench and in a laboratory environment which is completely different from an industrial scenario. The experimental machine is smaller than the ones used in the industries. The intelligence learning model trained on laboratory gearbox data will fail on the industrial gearbox.

[0018] And extracting all features of different failures like pitting, chipped tooth, broken tooth, corroded tooth, eroded tooth, foreign body, overheating, etc. Would not be possible to detect using one sensing element , and not feasible to install all possible sensing elements into the gearbox 12 because of very expensive cost, operating conditions, hazardous environmental conditions, and limited space at the gearbox housing itself. The above disclosed methodology helps in identifying/detecting the in-field fault in the gear box in the real-time efficiently by using the intelligence module that uses the pre-loaded data.

[0019] Data driven technique has taken an edge in identifying such faults in the gearbox 12. How ever gathering the faulty condition gear box data is key for efficient working of the gear box. The key features of the gear failures are mimicked in the numerical based simulation. Obtained healthy and non-healthy featured data were used in intelligence module methodologies to detect the faults in real-time operation of gearbox and identify the signatures of failure when they initiate.

[0020] Synthetic data generation is also a viable alternative to mitigate the lack of fault data to train the intelligence learning model in the intelligence module 20, in the field of condition-based maintenance for industrial machinery. Multibody dynamic simulations are used to create a simulation of the gearbox and generate data under healthy and various faulty conditions.

[0021] Multibody dynamic simulation can be used to model the behavior of a gearbox 12 under different operating conditions. The simulation would involve creating a digital model of the gearbox 12 and the numerical model is simulated from the interactions between the gearbox components, such as gears, shafts, bearings, prime mover, load, and lubricants. The simulation can be used to generate data on the gearbox under different loads, speeds, and temperatures, with healthy and faulty gears. The simulation model is used to generate synthetic vibration data for testing . These simulations closely model the real machine. The simulation model can then synthesize data for different fault conditions for the gearbox in large volumes and trains a neural network in the intelligence module. The model thus trained will have knowledge about various faults in a gearbox of a similar configuration to the simulation model. In the real time operating conditions, with the help of the above disclosed model, the intelligence module of the control unit 10 detects the in-field faults in the gear box 12.

[0022] With the above disclosed methodology, the faults can be detected at an early stage and the health of the machine can be accurately identified during the operating conditions. It provides a low-cost effective solution.

[0023] It should be understood that embodiments explained in the description above are only illustrative and do not limit the scope of this invention. Many such embodiments and other modifications and changes in the embodiment explained in the description are envisaged. The scope of the invention is only limited by the scope of the claims.
, Claims:We Claim:
1. A control unit (10) to detect at least one infield fault in an industrial machine (12), said machine (12) comprising :
- a sensing element (14) adapted to detect at least one machine parameter;
characterized in that :
said control unit (10) adapted to :
- develop a virtual model (16) of said machine (12) and build a virtual numerical model (18) of said machine (12) incorporating multiple real-time environments;
- generate data related to operating conditions of said machine (12) working by varying said at least one machine parameter;
- acquire different synthetic data related to said operating conditions of said machine (12) upon varying said at least one machine parameter;
- build and train an intelligence model (20) using said acquired different synthetic data in an intelligence module (20);
- detect said at least one infield fault in said machine (12) in a real-time using said intelligence module (20).

2. The control unit (10) as claimed in claim 1, wherein said at least one machine parameter detected by said sensing element (14) is any one of parameter chosen from a current, a number of vibrations in a given time period, a speed of said machine (12), a working temperature of said machine (12) .

3. The control unit (10) as claimed in claim 1, wherein said operating conditions of said machine (12) is chosen from the conditions wherein said machine (12) is working in different speeds, temperatures, loads.

4. The control unit (10) as claimed in claim 1, wherein acquisition of
said different synthetic data comprises both a healthy operating condition and a faulty operating condition of said machine (12).

5. The control unit (10) as claimed in claim 1, wherein said different synthetic data is pre-loaded into the intelligence module (18) for real-time detection of the working state of said machine (12).

6. The control unit (10) as claimed in claim 1, wherein said machine is a gear box (12).
7. The control unit (10) as claimed in claim 1, wherein said sensing element (14) adapted to detect a number of vibrations of said gear box (12).
8. A method for detecting at least one infield fault in an industrial machine (12), said method comprising :
- detecting at least one machine parameter of said machine (12) by a sensing element (14) ;
characterized in that :
- developing a virtual model (16) of said machine (12) and building a virtual numerical model (18) of said machine (12) incorporating multiple real-time environments by a control unit (10);
- generating data related to operating conditions of said machine (12) working by varying said at least one machine parameter;
- acquiring different synthetic data related to said operating conditions of said machine (12) upon varying at least one machine parameter;
- building and training an intelligence model using said acquired different synthetic data in an intelligence module (20);
- detecting said at least one infield fault in said machine (12) in a real-time using said intelligence module (20).