Method and device for diagnosing fault cause and computer storage medium

Abstract

The invention provides a method for diagnosing a fault cause. The method comprises the following steps: receiving a signal to be diagnosed; decomposing the to-be-diagnosed signal into a plurality of sub-signals; respectively converting each of the plurality of sub-signals into a corresponding gray scale image; and inputting the corresponding gray scale image into the neural network model, and outputting a fault cause classification result through the neural network model. Therefore, the method can be suitable for diagnosing fault reasons, and the problem that the fault reasons cannot be diagnosed is solved. In addition, a device for diagnosing the fault cause and a computer storage medium are also provided.

Description

Technical Field

[0001] The present application relates to a diagnostic method, device and computer storage medium, and specifically to a method, device and computer storage medium for diagnosing the cause of a fault. Background Art

[0002] With the booming development of Industry 4.0, many factories continue to provide automated equipment to improve factory production capacity. Most of these equipment are required to operate 24 hours a day without stopping, so it is very important to predict the maintenance needs of these equipment. Through the capture and analysis of some signals, if the maintenance time of these equipment can be predicted, or the different problem points of these equipment can be found, it will greatly increase the user's control over these equipment and help the operation and maintenance of production capacity.

[0003] Generally speaking, the fault analysis of these devices can be divided into manual analysis and automatic identification. However, since manual analysis requires a high degree of reliance on the judge's practical judgment experience, and it is difficult for the judge to pass on practical judgment experience to others, in terms of future trends, the fault analysis of these devices will inevitably tend to use artificial intelligence to replace manual analysis.

[0004] With regard to the prediction and judgment of these devices by collecting signals, the existing technologies can only predict that these devices are about to fail, but still cannot effectively distinguish the type of failure that these devices are about to fail, so it is impossible to effectively find the appropriate maintenance method (that is, it is impossible to effectively perform preventive maintenance for the upcoming problems). In other words, the existing technologies can only predict that these devices are about to fail, and then replace the motor or perform other basic maintenance for the devices that are about to fail. Summary of the invention

[0005] Based on the above problems, the purpose of this application is to solve the deficiencies of the prior art. Specifically, the purpose of this application is to solve the problem of being unable to predict the cause of a fault.

[0006] The present application provides a method for diagnosing the cause of a fault, the method comprising: receiving a signal to be diagnosed; decomposing the signal to be diagnosed into multiple sub-signals; converting each of the multiple sub-signals into a corresponding grayscale image; and inputting the corresponding grayscale image into a neural network model, and outputting a fault cause classification result through the neural network model.

[0007] In some embodiments, the method for diagnosing the cause of a fault provided in the present application may also include: comparing the number of multiple sub-signals with a predetermined number of grayscale images that the neural network model can receive; when the number of multiple sub-signals is greater than or equal to the predetermined number of grayscale images, stopping decomposing the signal to be diagnosed; and when the number of multiple sub-signals is less than the predetermined number of grayscale images, providing at least one filling image to the neural network model.

[0008] In addition, the present application also provides a device for diagnosing the cause of a fault, which is suitable for being connected to a device to be diagnosed signal to receive a signal to be diagnosed from the device to be diagnosed, and the device comprises: a signal receiving unit, which is suitable for receiving the signal to be diagnosed; a processing unit, which is configured to be coupled to the signal receiving unit; and a storage unit, which is configured to be coupled to the processing unit. The storage unit stores a code, and the processing unit can complete the method for diagnosing the cause of a fault as described above after executing the code stored in the storage unit.

[0009] In addition, the present application also provides a computer storage medium capable of diagnosing the cause of a fault. When the computer loads the code into the memory and executes it, the method for diagnosing the cause of a fault as previously described can be completed.

[0010] The technical solution provided by the present application can produce beneficial effects that cannot be achieved by the prior art. Specifically, the beneficial effect of the present application is to predict the cause of the fault so that the user can perform corresponding maintenance treatment according to each predicted fault cause. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a flow chart of a method for diagnosing the cause of a fault according to an embodiment of the present application.

[0012] Figure 2 It is a schematic diagram of an example of decomposing a signal to be diagnosed into multiple sub-signals.

[0013] Figure 3 is a detailed flowchart of an example of converting each of a plurality of sub-signals into a corresponding grayscale image.

[0014] Figure 4 The figure is a schematic diagram of an exemplary conversion result of converting each of the multiple sub-signals into a corresponding grayscale image.

[0015] Figure 5 It is a flowchart of a method for diagnosing a fault cause according to another embodiment of the present application.

[0016] Figure 6 It is a block diagram of the signal connection between the apparatus for diagnosing the cause of a fault and the device to be diagnosed according to one embodiment of the present application.

[0017] Reference numerals

[0018] 100 Device for diagnosing the cause of a fault

[0019] 110 Signal receiving unit

[0020] 120 processing units

[0021] 130 storage units

[0022] 200 Devices to be diagnosed

[0023] 300 Signal Line

[0024] C1~C8 sub-signal

[0025] GFPN Grayscale Image Pre-order Quantity

[0026] S110, S120, S130 steps

[0027] S130A First Step

[0028] S130B Second step

[0029] S130C Step 3

[0030] S130D Step 4

[0031] S140 Steps

[0032] Steps S210, S220, S230

[0033] Number of SSN sub-signals

[0034] u Signal to be diagnosed DETAILED DESCRIPTION

[0035] The present application will explain the contents of the present application in detail through the following embodiments and drawings, so as to help those skilled in the art understand the purpose, features and effects of the present application. It should be noted that the various steps described herein can be performed sequentially, in reverse order, or by appropriately changing or skipping the order during the control process.

[0036] Before the present application is described in detail, it should be noted that in the following description, the same elements or steps may be represented by the same reference numerals.

[0037] Please refer to Figure 1 , Figure 1 It is a flowchart of a method for diagnosing a fault cause according to an embodiment of the present application, and the method comprises steps S110, S120, S130 and S140.

[0038] In step S110, a signal to be diagnosed is received. The signal to be diagnosed is a signal generated by the device to be diagnosed, that is, step S110 can be connected to the device to be diagnosed through a signal line known to those skilled in the art to receive the signal to be diagnosed.

[0039] In step S120, the signal to be diagnosed is decomposed into a plurality of sub-signals, which can be specifically decomposed by methods such as wavelet transform, steady-state wavelet transform or empirical mode decomposition (EMD).

[0040] Since EMD can be used to process various types of signals, including nonlinear signals and non-steady-state signals, such as stock market indexes, electrocardiograms, voice signals or vibration signals, etc. Furthermore, EMD does not require any assumptions or prior knowledge about the signal, so that EMD will not cause distortion of the analysis results due to inaccurate assumptions. In addition, EMD can better capture the detailed information in the signal by decomposing the signal into multiple local components, making EMD more effective when extracting local features of the signal. Therefore, in some embodiments, the signal to be diagnosed can be decomposed into multiple sub-signals through the empirical mode decomposition model, so that various complex signals can be decomposed into a limited number of intrinsic mode functions (IMF).

[0041] In some embodiments, the decomposition process of decomposing the signal to be diagnosed into multiple sub-signals by using the empirical mode decomposition model may include the following steps:

[0042] The first step is to find the upper envelope u(t) and the lower envelope l(t) from the input signal S(t). In other words, the first step is to find all the local maximum points and local minimum points from the signal to be diagnosed S(t), and use the cubic spline interpolation function to fit the upper envelope u(t) and the lower envelope l(t) of the signal to be diagnosed S(t).

[0043] The second step is to calculate the average line m(t) of the upper envelope u(t) and the lower envelope l(t), and calculate the difference between the input signal S(t) and the average line m(t) to obtain a new signal h(t). In other words, the second step will calculate the average line m(t) of the upper envelope u(t) and the lower envelope l(t) of the signal to be diagnosed S(t), and calculate the difference between the signal to be diagnosed S(t) and the average line m(t) to obtain a difference line h(t).

[0044] The third step is to verify the new signal h(t) through the eigenmode function condition, where the eigenmode function condition can be:

[0045]

[0046] or

[0047] |local_max_num-local_min_mum|≤1

[0048] If the difference line h(t) cannot satisfy the above mentioned eigenmode function condition, the difference line h(t) is used to replace the signal to be diagnosed S(t), and the first step is performed again. k (t) = h k-1 (t)-m k When (t) satisfies the above mentioned eigenmode function conditions, the first eigenmode function will be extracted, i.e., c1(t) = h k (t). An eigenmode function must satisfy the following two conditions: (1) the number of local extrema and zero crossings must be equal or differ by at most one over the entire time range of the function; and (2) at any time point, the average of the envelope of the local maximum (upper envelope) and the envelope of the local minimum (lower envelope) must be zero.

[0049] The fourth step is to subtract the first intrinsic mode function c1(t) obtained in the third step from the input signal S(t) to retain the residual signal R1(t), that is, R1(t)=S(t)-c1(t). In other words, the fourth step is to subtract the first intrinsic mode function c1(t) obtained in the third step from the signal to be diagnosed S(t) to retain the residual signal R1(t).

[0050] The fifth step is to determine whether the residual signal R1(t) is a monotonic function. If the residual signal R1(t) is not a monotonic function, the residual signal R1(t) is used as the input signal S(t) and the first to fourth steps are executed again. If the residual signal R1(t) is a monotonic function, the sixth step is executed.

[0051] The sixth step is to obtain n eigenmode functions c1(t) to c n (t), therefore, the input signal S(t) can be expressed as:

[0052]

[0053] The remaining signal will be regarded as the residual signal.

[0054] Please refer to Figure 2 , Figure 2 The source is: Norden E.Huang, et al. "The empirical modedecomposition and the Hilbert spectrum for nonlinear and non-stationary timeseries analysis." Proc.R.Soc.Lond.A (1998) 454,903–995, Figure 2is a schematic diagram of an example of decomposing a signal to be diagnosed into multiple sub-signals, which is specifically a decomposition result of an example after executing step S120 through the empirical mode decomposition model as described above. That is, after the signal to be diagnosed u executes step S120 as described above through the empirical mode decomposition model, multiple sub-signals C1 to C8 can be obtained, wherein the multiple sub-signals C1 to C8 can also be referred to as eigenmode functions c1(t) to c8(t).

[0055] By executing step S120 through the empirical mode decomposition model, a complex signal can be decomposed into a limited number of intrinsic mode functions, and each of the decomposed intrinsic mode functions contains local characteristic signals of different time scales of the signal to be diagnosed. Therefore, the empirical mode decomposition model can decompose various types of complex signals, which helps to perform more effective signal processing and analysis on the decomposed signal (i.e., the intrinsic mode function) in the subsequent process, such as further converting the intrinsic mode function into a Hilbert spectrum.

[0056] In step S130 , each of the plurality of sub-signals is converted into a corresponding grayscale image.

[0057] Please refer to Figure 3 , Figure 3 The source is: https: / / www.cs.brandeis.edu / ~cs136a / CS136a_docs / KishorePrahallad_CMU_mfcc.pdf (Kishore Prahallad's slides), Figure 3 is a detailed flow chart of an example of converting each of the plurality of sub-signals into a corresponding grayscale image. In some embodiments, the conversion process of converting the sub-signals into the corresponding grayscale images may include the following steps:

[0058] The first step S130A is to divide the sub-signal into a plurality of signal segments, for example, to divide the first eigenmode function c1(t) into a plurality of first signal segments. At this time, the plurality of signal segments are still represented in the time domain.

[0059] The second step S130B is to transform the multiple signal segments from the time domain signal into the frequency domain signal, for example, each of the multiple first signal segments of the first eigenmode function c1(t) is transformed from the time domain signal into the frequency domain signal. In some embodiments, the conversion can be performed by a Fourier transform model such as a Fast Fourier Transform (FFT).

[0060] The third step S130C is to map the signal segments in the form of multiple frequency domain signals to corresponding grayscale values ​​based on the signal strength (i.e., amplitude value) of the signal segments. In some embodiments, the grayscale value may be between 0 and 255, wherein a grayscale value of 0 may represent black, and a grayscale value of 255 may represent white, but is not limited thereto. In one example, a range of signal strength may be set (e.g., 15 volts to -15 volts, but is not limited thereto), and then each signal segment may be normalized based on the range of signal strength to be mapped to corresponding grayscale values ​​respectively.

[0061] The fourth step S130D is to integrate each grayscale value (ie, the grayscale value corresponding to each signal segment) into a two-dimensional grayscale matrix, and then convert it into a corresponding grayscale image based on the two-dimensional grayscale matrix.

[0062] Please refer to Figure 4 , Figure 4 A schematic diagram of a conversion result of an example of converting each of the plurality of sub-signals into a corresponding grayscale image. Through step S130, each of the plurality of sub-signals can be mapped to a corresponding grayscale value to obtain a two-dimensional grayscale matrix, and then generate a corresponding grayscale image. In some embodiments, the grayscale image can be a two-dimensional grayscale image that can represent each grayscale value, but is not limited thereto.

[0063] In step S140, the corresponding grayscale image is input into the neural network model, and the fault cause classification result is output through the neural network model. The neural network model may refer to a neural network model that has been trained through multiple data, wherein each of the multiple data may include a machine system signal and a machine system state corresponding to the machine system signal, and the structure of the neural network model may be a structure known to those skilled in the art. In some embodiments, the neural network model may be a convolutional neural network model or a recursive neural network model, but is not limited thereto. Among them, the convolutional neural network model may be, for example, PyTorch RESNET, but is not limited thereto.

[0064] Through Figure 1 The method for diagnosing the cause of a fault shown can diagnose the device to be diagnosed to generate a fault cause classification result, thereby knowing the type of fault that may be about to occur in the device to be diagnosed. In some embodiments, the fault cause classification result can include at least one of imbalance, non-parallelism, shaft deflection, and looseness, but is not limited thereto.

[0065] Please refer to Figure 5 , Figure 5is a flowchart of a method for diagnosing a fault cause according to another embodiment of the present invention, the method comprising steps S110, S120, S130, S140, S210, S220 and S230, wherein steps S110, S120, S130 and S140 are Figure 1 The steps shown are basically the same. That is, Figure 5 The method shown may include Figure 1 The steps S110, S120, S130 and S140 are basically the same, and further include steps S210, S220 and S230.

[0066] In step S210, the number of sub-signals SSN is compared with the predetermined number of grayscale images GFPN. When the number of sub-signals SSN is less than the predetermined number of grayscale images GFPN, at least one filling image is provided to the neural network model (i.e., step S220), and step S130 is continued. When the number of sub-signals SSN is greater than or equal to the predetermined number of grayscale images GFPN, the decomposition of the signal to be diagnosed is stopped (i.e., step S230), and step S130 is continued. In some embodiments, the predetermined number of grayscale images GFPN can be preset to 8, but is not limited thereto.

[0067] In some embodiments, the number of filling images in step S220 may depend on the number of sub-signals SSN and the predetermined number of grayscale images GFPN. For example, the number of filling images may be the difference between the predetermined number of grayscale images GFPN and the number of sub-signals SSN. For example, when the predetermined number of grayscale images GFPN is preset to 8, if the number of sub-signals SSN is 5, the number of filling images is 3.

[0068] Since the number of sub-signals decomposed in step S120 depends on the signal to be diagnosed, Figure 5 The method for diagnosing the cause of a fault shown can input a fixed number of images (the total number of grayscale images and fill images is equal to the predetermined number of grayscale images) into a neural network model, so that the neural network model can predict the fault cause classification results of the equipment to be diagnosed based on the same number of images.

[0069] [Example 1]

[0070] Pre-training of neural network models

[0071] 1. Use PyTorch RESNET (https: / / pytorch.org / hub / pytorch_vision_resnet / ) to build an image classifier.

[0072] 2. Assume that the EMD result is less than or equal to 10 (i.e., the number of multiple sub-signals decomposed by the empirical mode decomposition model is set to be less than or equal to 10), the IMF is split into frequency spectrum diagrams in units of 100 milliseconds (i.e., the first step S130A decomposes each intrinsic mode signal into multiple signal segments in units of 100 milliseconds), and the amplitude value is between -15 and 15 (i.e., the third step S130C normalizes the multiple signal segments based on the amplitude values ​​to map them to corresponding grayscale values).

[0073] 3. Therefore, the shape and size of the image can be the number of EMDs × the amplitude value × the number of multiple signal segments. If the total receiving time of the diagnosed signal is 100 seconds, the calculation formula is 1000000 / 100=1000 (that is, the shape and size of the image is 10×30×1000), and the image size is adjusted from 1000 to the default value (that is, 224), that is, 10×30×224.

[0074] By using the neural network model as described above and pre-training with multiple data, a trained neural network model can be generated (for example, a neural network model with a recognition accuracy higher than 0.995, but not limited thereto). Thus, the trained neural network model can be used to diagnose the device to be diagnosed, so as to generate a classification result of the cause of the fault, and then the type of fault that may be about to occur in the device to be diagnosed can be known.

[0075] Please refer to Figure 6 , Figure 6 The block diagram of the signal connection between the apparatus 100 for diagnosing the cause of a fault and the device 200 to be diagnosed according to an embodiment of the present application is shown in FIG. The apparatus 100 for diagnosing the cause of a fault may include a signal receiving unit 110 , a processing unit 120 , and a storage unit 130 .

[0076] The apparatus 100 for diagnosing the cause of a fault may be connected to the device 200 for diagnosis by signal, which may be specifically connected by signal lines such as RS-232 or RS-485. Therefore, the signal receiving unit 110 in the apparatus 100 for diagnosing the cause of a fault may receive the signal to be diagnosed from the device 200 for diagnosis via the signal line 300.

[0077] The processing unit 120 may be a product known to those skilled in the art, and may specifically be composed of one or more central processing units, but is not limited thereto. The processing unit 120 is configured to be coupled to the signal receiving unit 110, and is used to process the signal to be diagnosed received by the signal receiving unit 110, and the processing may refer to executing the various steps of the method for diagnosing the cause of the fault as described above. That is, after executing a specific code or instruction set, the processing unit 120 can complete the method for diagnosing the cause of the fault as described above based on the signal to be diagnosed received by the signal receiving unit 110.

[0078] The storage unit 130 may be a finished product known to those skilled in the art, and may specifically be composed of a volatile memory and a non-volatile memory, but is not limited thereto. The volatile memory is a finished product known to those skilled in the art, such as a dynamic random access memory or a static random access memory, but is not limited thereto; the non-volatile memory may be a finished product known to those skilled in the art, such as a read-only memory, a flash memory, or a non-volatile static random access memory, but is not limited thereto. The storage unit 130 stores a code or an instruction set, and is configured to be coupled to the processing unit 120, so that the processing unit 120 can execute the code or instruction set stored in the storage unit 130, so as to complete the method for diagnosing the cause of the fault as previously described based on the signal to be diagnosed received by the signal receiving unit 110.

[0079] Through Figure 6 The device 100 for diagnosing the cause of a fault as shown can be connected to the device to be diagnosed 200 by signal, and diagnose the device to be diagnosed based on the signal to be diagnosed from the device to be diagnosed to generate a fault cause classification result, and then know the type of fault that may be about to occur in the device to be diagnosed.

[0080] In addition, in some embodiments, the steps of the method for diagnosing the cause of a fault as described above may be stored in a computer storage medium, which may be, for example, hardware, an optical disk, a magnetic disk, a universal serial bus, or a database accessible by the cloud, but is not limited thereto. The computer storage medium can complete the method for diagnosing the cause of a fault as described above after the code or instruction set loaded into the memory and executed by a computer, a computer device, or other device capable of executing code.

[0081] Since the computer-readable recording medium stores codes or instruction sets that can complete the method of diagnosing the cause of a fault as described previously, a computer, a computer device, or other device capable of executing codes or instruction sets can diagnose the device to be diagnosed after loading and executing the codes or instruction sets to generate a fault cause classification result, thereby determining the type of fault that may be about to occur in the device to be diagnosed.

[0082] The present application has been further described through the above embodiments and drawings, but those skilled in the art can still make many modifications and changes without violating the scope and spirit of the claims of the present application. Therefore, the protection scope of the present application shall still be based on the content defined in the claims and shall not be limited by the content disclosed in the specification.

Claims

1. A method for diagnosing the cause of a fault, characterized in that: Include: receiving a signal to be diagnosed; Decomposing the signal to be diagnosed into a plurality of sub-signals; converting each of the plurality of sub-signals into a corresponding grayscale image; and A plurality of corresponding grayscale images are input into a neural network model, and a fault cause classification result is output through the neural network model.

2. The method according to claim 1, characterized in that The decomposition of the signal to be diagnosed into multiple sub-signals is accomplished by an empirical mode decomposition model.

3. The method according to claim 1, characterized in that The plurality of sub-signals are eigenmode functions.

4. The method according to claim 1, characterized in that: The converting each of the plurality of sub-signals into the corresponding grayscale image comprises: Performing signal transformation on the multiple sub-signals through a Fourier transform model; Mapping the signal strengths of the plurality of sub-signals after signal transformation into corresponding grayscale values ​​respectively; and A plurality of corresponding grayscale values ​​are integrated into the corresponding grayscale image.

5. The method according to claim 1, characterized in that Also includes: Comparing the number of the plurality of sub-signals with a predetermined number of grayscale images that can be received by the neural network model; When the number of the plurality of sub-signals is greater than or equal to the predetermined number of the grayscale images, stopping decomposing the signal to be diagnosed; as well as When the number of the plurality of sub-signals is less than a predetermined number of the grayscale image, at least one filling image is provided to the neural network model.

6. The method according to claim 5, characterized in that The number of the at least one filling image depends on the number of the plurality of sub-signals and the predetermined number of the grayscale images.

7. The method according to claim 1, characterized in that The neural network model is a convolutional neural network model trained by multiple data, and each of the multiple data includes a machine system signal and a machine system state corresponding to the machine system signal.

8. The method according to claim 1, characterized in that The fault cause classification result includes at least one of imbalance, non-parallelism, shaft deflection and looseness.

9. A device for diagnosing the cause of a fault, adapted to be connected to a device to be diagnosed by signal, so as to receive a signal to be diagnosed from the device to be diagnosed, characterized in that: The device comprises: A signal receiving unit, adapted to receive the signal to be diagnosed; a processing unit configured to be coupled to the signal receiving unit; and a storage unit configured to be coupled to the processing unit, wherein the storage unit stores a code, and After executing the code stored in the storage unit, the processing unit can execute the method for diagnosing the cause of a fault according to any one of claims 1 to 8.

10. A computer storage medium capable of diagnosing the cause of a fault, which can complete the method for diagnosing the cause of a fault according to any one of claims 1 to 8 after the computer loads the code into the memory and executes it.

Images