Intelligent fault diagnosis method and system for multiple components of low-rotating-speed mechanical transmission system

Abstract

The invention discloses a multi-component intelligent fault diagnosis method and system for a low-rotating-speed mechanical transmission system, belongs to the field of mechanical fault diagnosis, and adopts improved robust local binarization operation to convert a vibration signal into a feature enhanced gray texture image to highlight component feature information in original vibration data. A vibration gray texture image noise reduction model based on an auto-encoder is constructed, and the model has a good noise reduction effect on different degrees of noise. By optimizing the width of the convolutional neural network model, the light weight of the fault diagnosis model is realized, and the fault diagnosis efficiency is improved while the fault diagnosis accuracy is ensured. According to the method, the problem of low accuracy of a fault diagnosis model caused by weak vibration signal energy and noise interference of a traditional mechanical system in a low-rotating-speed and noise environment is solved, and an effective solution is provided for intelligent fault diagnosis of each part of the mechanical transmission system in the low-rotating-speed and noise environment.

Description

Technical Field

[0001] The present invention belongs to the field of mechanical fault diagnosis, relates to the field of mechanical transmission system fault diagnosis, and specifically relates to a multi-component intelligent fault diagnosis method and system for a low-speed mechanical transmission system. Background Art

[0002] The mechanical transmission system is a key component of mechanical equipment and is a complex system consisting of many parts. Among them, bearings, rotors and gears play the most critical role in the process of power transmission and motion conversion and are the most vulnerable parts in mechanical equipment. Therefore, fault diagnosis of the mechanical transmission system composed of rotors, bearings and gears is crucial to the safe and reliable operation of mechanical equipment.

[0003] For a long time, fault diagnosis of mechanical transmission systems has mainly used vibration signals, extracted fault features through deep signal analysis, and then completed fault diagnosis through expert experience. The fault diagnosis ability of fault diagnosis methods based on signal analysis is limited, and the effect of use is heavily dependent on manual experience, and it is impossible to achieve automatic diagnosis of various faults without human intervention. With the rapid development of big data analysis and artificial intelligence technology, deep learning has gradually been used in the field of mechanical fault diagnosis due to its advantages of automatically acquiring data features and automatically classifying data.

[0004] In engineering practice, the vibration signals of mechanical equipment are often affected by strong background noise. Mechanical fault diagnosis methods lose the accuracy of fault identification when diagnosing noisy vibration signals. In addition, the effect of mechanical fault diagnosis methods based on deep learning is difficult to adapt to the data being analyzed. It needs to be adjusted autonomously according to different research purposes and data distribution patterns. Overly complex deep learning models are difficult to train and difficult to apply under low-resource conditions. When the vibration signal is converted into a two-dimensional image, the problems of the above-mentioned deep learning-based fault diagnosis method will be aggravated. It is urgent to design a deep learning fault diagnosis method suitable for vibration images.

[0005] In addition, the vibration signal does not show obvious regularity at low speeds, because the vibration regularity is affected by nonlinear characteristics due to the existence of gaps between components in the mechanical transmission system at low speeds, making the distribution regularity of vibration signals in different fault states similar, resulting in a decrease in the classification and recognition ability of deep learning-based methods for such data. Moreover, in the mechanical transmission system, different parts have different sensitivities to different forms of vibration signals. The rotor and the inner ring of the rolling bearing are more sensitive to vibration displacement signals and are suitable for monitoring using eddy current sensors, while the gears and non-inner ring parts of the rolling bearing are more sensitive to vibration acceleration and are suitable for monitoring using piezoelectric sensors. Therefore, different types of vibration signals are needed to monitor the status of different parts of the mechanical transmission system, so as to achieve the best fault diagnosis. Summary of the invention

[0006] The purpose of the present invention is to provide a multi-component intelligent fault diagnosis method and system for a low-speed mechanical transmission system, so as to overcome the problem that the mechanical equipment is in a low-speed and strong noise environment, resulting in unclear visualization of vibration signals and ineffective intelligent fault diagnosis methods. The present invention designs a method for converting vibration signals into grayscale texture images under low-speed conditions, combines an autoencoder vibration grayscale texture image denoising model and a lightweight convolutional neural network intelligent diagnosis model, and realizes efficient, accurate and intelligent diagnosis of multi-component faults of mechanical transmission systems under low-speed and noise conditions.

[0007] In order to achieve the above object, the present invention adopts the following technical scheme: A multi-component intelligent fault diagnosis method for a low-speed mechanical transmission system comprises the following steps: 1) Constructing a mechanical transmission system vibration grayscale texture image dataset The vibration acceleration and displacement signals of bearings, rotors and gears in the mechanical transmission system under normal and fault conditions at low speed are collected. The vibration acceleration and displacement signals are segmented and spliced ​​into feature matrices according to the specified number of points, and then standardized, proportionally scaled and robustly locally binarized are processed in sequence to generate the mechanical transmission system vibration grayscale texture image. All the generated images constitute the mechanical transmission system vibration grayscale texture image dataset, and the vibration grayscale texture image dataset is divided into a training set and a test set in proportion. 2) Construct an autoencoder vibration grayscale texture image denoising model Based on the mechanical transmission system vibration grayscale texture image dataset in step 1), a denoising model of the autoencoder vibration grayscale texture image is constructed, wherein the denoising model comprises multi-scale convolution, Ghost convolution and ECB convolution, the training set is used for model training, the test set is used for model testing, and the denoising effect of the denoising model is evaluated by peak signal-to-noise ratio and structural similarity indicators; 3) Construct a fault diagnosis model for mechanical transmission systems Based on the lightweight convolutional neural network, the model width is optimized to construct an improved convolutional neural network fault diagnosis model. Based on the mechanical transmission system vibration grayscale texture image dataset in step 1), a fault feature dataset is established after the fault label is added, and it is divided into a training set and a test set. The training set is used for model training, and the test set verifies the final diagnosis result. The final structure and parameters of the fault diagnosis model are saved. After the model structure and parameters are determined, three different degrees of noise, light, medium and heavy, are added to the mechanical transmission system vibration grayscale texture image dataset in step 1) to establish a noisy vibration grayscale texture image dataset. According to the evaluation index of the denoising model in step 2), the optimal model parameters are obtained, and the optimal parameters are used to denoise the vibration grayscale texture image datasets with different degrees of noise. After the denoised vibration grayscale texture images are labeled, a denoised fault feature dataset is established, and it is divided into a training set and a test set. The training set is used for model training, and the test set verifies the diagnostic effect of the fault diagnosis model on the denoised image dataset.

[0008] Furthermore, the step 1) constructs a mechanical transmission system vibration grayscale texture image dataset, specifically: 1.1) Collect the vibration acceleration and displacement signals of bearings, rotors and gears in the mechanical transmission system under normal and fault conditions at low speed, and segment and splice the vibration acceleration and displacement signals according to the specified number of points to form a feature matrix; 1.2) The elements of the feature matrix are normalized from 0 to 255 and converted into a grayscale image. The conversion relationship between the feature matrix element value and the grayscale image pixel value is:

[0009] In the formula, i , j Represents the position coordinates in the feature matrix or grayscale image, is the feature matrix located at ( i , j ) element, X max and X min Represent the maximum and minimum values ​​in the feature matrix respectively, is the gray value of the corresponding position after transformation; 1.3) Scale the converted grayscale image to 256×256 pixels; 1.4) Perform robust local binarization on the image obtained after equal-proportion scaling to generate a vibration grayscale texture image, and introduce weighted local grayscale values ​​in the robust local binarization process to adjust the weight of the center pixel for threshold selection;

[0010] In the formula, is the gray value of the center pixel, The radius is R The grayscale values ​​of adjacent pixels on the circle, is a constant, The larger the value, the greater the proportion of the center threshold; After completing the threshold selection, robust local binarization is performed to describe the texture features of the image. The size of the circular pixel area in the image is calculated in sequence. The calculation formula for robust local binarization is:

[0011] In the formula, p Represents the total number of selected area points on the circumference, , Represents the gray value of the area point, RLBP P,R is the RLBP value of the center point after being processed by the RLBP operator, ; 1.5) Add different degrees of Gaussian noise to all the obtained vibration grayscale texture images to construct a noisy grayscale texture image dataset, and divide it into a training set and a test set according to a preset ratio.

[0012] Furthermore, the preset ratio is 8:2.

[0013] Furthermore, the generated image samples contain prominent texture detail features, and the autoencoder denoising model is used to perform feature extraction and image reconstruction on the noisy vibration grayscale texture image dataset in step 1.5) to achieve the purpose of denoising.

[0014] Furthermore, in step 2), an autoencoder vibration grayscale texture image denoising model is constructed, specifically: 2.1) Construct an autoencoder vibration grayscale texture image denoising model, including multi-scale convolution, Ghost convolution and ECB convolution; 2.2) Use step 1) to construct a vibration grayscale texture image dataset of bearings, rotors, and gears in mechanical transmission systems under normal and faulty conditions at low speed. Divide the training set and test set into a ratio of 8:2, add three different degrees of Gaussian noise with standard deviations of 15, 25, and 50 to the dataset, and achieve the optimal denoising effect by updating the model hyperparameter settings. The denoising effect of the denoising model is evaluated by the peak signal-to-noise ratio and structural similarity indicators.

[0015] Furthermore, in step 3), a mechanical transmission system fault diagnosis model is constructed, specifically: 3.1) Construct a convolutional neural network fault diagnosis model, including a feature extractor and a classifier, optimize the width of the model structure of the feature extraction part, and simplify the model structure; 3.2) For low speed working conditions, step 1) is used to construct a vibration grayscale texture image dataset of bearings, rotors, and gears in the mechanical transmission system under normal and faulty conditions. The vibration grayscale texture images after denoising in step 2) are labeled to establish a dataset, which is divided into a training set and a test set. The dataset is input into the fault diagnosis model for training and testing. The classification accuracy, precision, recall rate, and F1 score of the fault diagnosis model on the test set are used as evaluation indicators. The calculation formula is:

[0016]

[0017]

[0018]

[0019] Where ACC, precision, recall and F1 represent the accuracy, precision, recall and F1 score of the test set respectively; TP represents the number of samples that are actually positive and correctly predicted as positive, TN represents the number of samples that are actually negative and correctly predicted as negative; FP represents the number of samples that are actually negative but incorrectly predicted as positive; FN represents the number of samples that are actually positive but incorrectly predicted as negative.

[0020] A multi-component intelligent fault diagnosis system for a low-speed mechanical transmission system, comprising: Mechanical transmission system vibration grayscale texture image dataset construction module: collect the vibration acceleration and displacement signals of bearings, rotors, and gears in the mechanical transmission system under normal and faulty conditions at low speed, segment and splice the vibration acceleration and displacement signals into feature matrices according to the specified number of points, and then perform standardization, proportional scaling, and robust local binarization processing in sequence to generate the corresponding vibration grayscale texture images. All the generated images constitute the mechanical transmission system vibration grayscale texture image dataset, and the vibration grayscale texture image dataset is divided into a training set and a test set in proportion; An autoencoder vibration grayscale texture image denoising model construction module is used to use low-speed working condition data to construct a vibration grayscale texture image dataset of low-speed working condition through operations in a mechanical transmission system vibration grayscale texture image dataset construction module, and to construct an autoencoder vibration grayscale texture image denoising model using the vibration grayscale texture image dataset of low-speed working condition. The autoencoder denoising model includes multi-scale convolution, Ghost convolution and ECB convolution. The training set is used for model training, and the test set is used for model testing. The denoising effect of the denoising model is evaluated by peak signal-to-noise ratio and structural similarity indicators. Mechanical transmission system fault diagnosis model construction module: Construct a convolutional neural network fault diagnosis model, including a feature extractor and a classifier, optimize the width of the model structure of the feature extraction part, and simplify the model structure. For low-speed working condition data, build a vibration grayscale texture image dataset of bearings, rotors, and gears in the mechanical transmission system under normal and fault conditions, label the vibration grayscale texture image samples after denoising by the denoising model to establish a fault feature dataset, and divide it into a training set and a test set to input into the fault diagnosis model for training and testing. The classification accuracy, precision, recall rate, and F1 score of the fault diagnosis model on the test set are used as evaluation indicators to complete the fault diagnosis of the mechanical transmission system under low-speed working conditions and noise interference.

[0021] Furthermore, the construction process of the mechanical transmission system vibration grayscale texture image dataset is specifically as follows: 1.1) Collect the vibration acceleration and displacement signals of bearings, rotors and gears in the mechanical transmission system under normal and fault conditions at low speed, and segment and splice the vibration acceleration and displacement signals according to the specified number of points to form a feature matrix; 1.2) The elements of the feature matrix are normalized from 0 to 255 and converted into a grayscale image. The conversion relationship between the feature matrix element value and the grayscale image pixel value is:

[0022] In the formula, i , j Represents the position coordinates in the feature matrix or grayscale image, is the feature matrix located at ( i , j ) element, X max and X min represents the maximum and minimum values ​​in the feature matrix, respectively. is the gray value of the corresponding position after transformation; 1.3) Scale the converted grayscale image to 256×256 pixels; 1.4) Perform robust local binarization on the image obtained after equal-proportion scaling to generate a vibration grayscale texture image, and introduce weighted local grayscale values ​​in the robust local binarization process to adjust the weight of the center pixel for threshold selection;

[0023] In the formula, is the gray value of the center pixel, The radius is R The grayscale values ​​of adjacent pixels on the circle, is a constant, The larger the value, the greater the proportion of the center threshold; After completing the threshold selection, robust local binarization is performed to describe the texture features of the image. The size of the circular pixel area in the image is calculated in sequence. The calculation formula for robust local binarization is:

[0024] In the formula, p Represents the total number of selected area points on the circumference, , Represents the gray value of the area point, RLBP P,R is the RLBP value of the center point after being processed by the RLBP operator, ; 1.5) Add different degrees of Gaussian noise to all the obtained vibration grayscale texture images to construct a noisy grayscale texture image dataset, and divide it into a training set and a test set according to a preset ratio.

[0025] Furthermore, the construction of the autoencoder vibration grayscale texture image denoising model is specifically as follows: 2.1) Construct an autoencoder vibration grayscale texture image denoising model, including multi-scale convolution, Ghost convolution and ECB convolution; 2.2) Use step 1) to construct a vibration grayscale texture image dataset of bearings, rotors, and gears in mechanical transmission systems under normal and faulty conditions at low speed. Divide the training set and test set into a ratio of 8:2, add three different degrees of Gaussian noise with standard deviations of 15, 25, and 50 to the dataset, and achieve the optimal denoising effect by updating the model hyperparameter settings. The denoising effect of the denoising model is evaluated by the peak signal-to-noise ratio and structural similarity indicators.

[0026] Furthermore, the construction of the mechanical transmission system fault diagnosis model is specifically as follows: 3.1) Construct a convolutional neural network fault diagnosis model, including a feature extractor and a classifier, optimize the width of the model structure of the feature extraction part, and simplify the model structure; 3.2) For low speed working conditions, step 1) is used to construct a vibration grayscale texture image dataset of bearings, rotors, and gears in the mechanical transmission system under normal and faulty conditions. The vibration grayscale texture images after denoising in step 2) are labeled to establish a dataset, which is divided into a training set and a test set. The dataset is input into the fault diagnosis model for training and testing. The classification accuracy, precision, recall rate, and F1 score of the fault diagnosis model on the test set are used as evaluation indicators. The calculation formula is:

[0027]

[0028]

[0029]

[0030] Where ACC, precision, recall and F1 represent the accuracy, precision, recall and F1 score of the test set respectively; TP represents the number of samples that are actually positive and correctly predicted as positive, TN represents the number of samples that are actually negative and correctly predicted as negative; FP represents the number of samples that are actually negative but incorrectly predicted as positive; FN represents the number of samples that are actually positive but incorrectly predicted as negative.

[0031] Compared with the prior art, the present invention has the following beneficial technical effects: (1) To address the problem that the fault vibration characteristics are not obvious under low-speed conditions, an improved grayscale texture image sample generation method is proposed. The texture features and detail information of the grayscale image are enhanced through robust local binarization operation, and the obtained grayscale texture image information is used to highlight the status characteristics of different components of the mechanical transmission system. Compared with the conventional vibration grayscale image conversion method, the improved vibration grayscale texture image has a larger difference between samples of different fault categories. At the same time, compared with the grayscale texture image generated by the traditional local binarization operation, it has clearer texture details, which is conducive to further feature extraction.

[0032] (2) In order to reduce the negative impact of background noise on fault diagnosis in actual working environments, a multi-module vibration grayscale texture image denoising model based on autoencoder is proposed. The multi-scale convolution, Ghost convolution and ECB convolution in Inception are used to improve the autoencoder, which enhances the feature extraction and image reconstruction capabilities of the vibration grayscale texture image denoising model. The improved autoencoder model has the advantages of fewer parameters, less computation, faster speed and smaller model. It can remove background noise of different degrees and better restore the texture details of vibration grayscale texture images.

[0033] (3) An improved convolutional neural network fault diagnosis model is proposed. By optimizing the bottleneck layer structure of the basic lightweight convolutional neural network, the number of model parameters, the amount of computation, and the model size are reduced, while ensuring a high fault diagnosis accuracy. The proposed method can still achieve good results under background noise, has strong robustness to noise, and maintains a high fault diagnosis accuracy for the denoised data set.

[0034] In order to solve the problem of low accuracy of fault diagnosis model caused by weak energy of vibration signal and large noise interference of traditional mechanical system under low speed condition, an intelligent fault diagnosis method based on vibration signal mapping grayscale texture image and autoencoder pre-training convolutional neural network is proposed. The improved grayscale texture image converted from vibration signal under low speed condition is used for autoencoder denoising and convolutional neural network fault diagnosis. The accuracy, precision, recall rate and F1 score of data classification are used as indicators to evaluate the effect of fault diagnosis model. Compared with other fault diagnosis methods of mechanical transmission system, the present invention can accurately extract image features and maintain high fault diagnosis accuracy under low speed and noisy environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The drawings in the specification are used to provide further understanding of the present invention and constitute a part of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0036] Figure 1 It is a flow chart of a multi-component intelligent fault diagnosis method for a low-speed mechanical transmission system; Figure 2 It is the vibration grayscale texture image obtained by converting the vibration signal when the components are in different fault states; Figure 3 A schematic diagram of the structure of the vibration grayscale texture image autoencoder denoising model constructed by the present invention; Figure 4 A schematic diagram of the structure of the improved convolutional neural network fault diagnosis model constructed by the present invention; Figure 5 It is the classification performance of the present invention on data sets containing different degrees of noise and denoised data sets. DETAILED DESCRIPTION

[0037] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0038] It should be understood that the numbers used in this way are interchangeable where appropriate, so that the embodiments of the invention described herein can be implemented in sequences other than those illustrated or described herein. In addition, the terms "comprises" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or inherent to these processes, methods, products, or apparatus.

[0039] Embodiment 1 A multi-component intelligent fault diagnosis method for a low-speed mechanical transmission system comprises the following steps: 1) Constructing a mechanical transmission system vibration grayscale texture image dataset: The vibration acceleration and displacement signals of bearings, rotors and gears in the mechanical transmission system under normal and fault conditions at low speed are collected. The vibration acceleration and displacement signals are segmented and spliced ​​into feature matrices according to the specified number of points, and then standardized, proportionally scaled and robustly locally binarized are processed in sequence to generate the mechanical transmission system vibration grayscale texture image. All the generated images constitute the mechanical transmission system vibration grayscale texture image dataset, and the vibration grayscale texture image dataset is divided into a training set and a test set in proportion. The specific steps for constructing the mechanical transmission system fault feature dataset in step 1) are as follows: 1.1) Collect the vibration acceleration and displacement signals of bearings, rotors and gears in the mechanical transmission system under normal and fault conditions at low speed, and segment and splice the vibration acceleration and displacement signals according to the specified number of points to form a feature matrix; 1.2) The elements of the feature matrix are normalized from 0 to 255 and converted into a grayscale image. The conversion relationship between the feature matrix element value and the grayscale image pixel value is:

[0040] In the formula, i , j Represents the position coordinates in the feature matrix or grayscale image, is the feature matrix located at ( i , j ) element, X max and X min Represent the maximum and minimum values ​​in the feature matrix respectively, is the gray value of the corresponding position after transformation; 1.3) Scale the converted grayscale image to 256×256 pixels; 1.4) Perform robust local binarization on the image obtained after equal-proportion scaling to generate a vibration grayscale texture image, and introduce weighted local grayscale values ​​in the robust local binarization process to adjust the weight of the center pixel for threshold selection;

[0041] In the formula, is the gray value of the center pixel, The radius is R The grayscale values ​​of adjacent pixels on the circle, is a constant, The larger the value, the greater the proportion of the center threshold; After completing the threshold selection, robust local binarization is performed to describe the texture features of the image. The size of the circular pixel area in the image is calculated in sequence. The calculation formula for robust local binarization is:

[0042] In the formula, p Represents the total number of selected area points on the circumference, , Represents the gray value of the area point. P,R is the RLBP value of the center point after being processed by the RLBP operator, ; 1.5) Add different degrees of Gaussian noise to all the obtained vibration grayscale texture images to construct a noisy grayscale texture image dataset, and divide it into a training set and a test set according to a preset ratio.

[0043] 2) Constructing an autoencoder vibration grayscale texture image denoising model: Based on the mechanical transmission system vibration grayscale texture image dataset in step 1), constructing an autoencoder vibration grayscale texture image denoising model, the denoising model includes multi-scale convolution, Ghost convolution and ECB convolution, the training set is used for model training, the test set is used for model testing, and the denoising effect of the denoising model is evaluated by peak signal-to-noise ratio and structural similarity indicators; The specific steps of building the autoencoder vibration grayscale texture image denoising model in step 2) are as follows: 2.1) Construct an autoencoder vibration grayscale texture image denoising model, including multi-scale convolution, Ghost convolution and ECB convolution.

[0044] 2.2) Use step 1) to construct a vibration grayscale texture image dataset of bearings, rotors, and gears in mechanical transmission systems under normal and faulty conditions at low speed. Divide the training set and test set into a ratio of 8:2, add three different degrees of Gaussian noise with standard deviations of 15, 25, and 50 to the dataset, and achieve the optimal denoising effect by updating the model hyperparameter settings. The denoising effect of the denoising model is evaluated by the peak signal-to-noise ratio and structural similarity indicators.

[0045] 3) Constructing a fault diagnosis model: MobileNetV3 is selected as the basic model of this embodiment, and the model width is optimized to construct a fault diagnosis model of an improved convolutional neural network. Based on the mechanical transmission system vibration grayscale texture image dataset in step 1), a fault feature dataset is established after the fault label is marked, and it is divided into a training set and a test set. The training set is used for model training, and the test set verifies the final diagnosis result. The final structure and parameters of the fault diagnosis model are saved. After the model structure and parameters are determined, three different degrees of noise, light, medium, and heavy, are added to the mechanical transmission system vibration grayscale texture image dataset in step 1) to establish a noisy vibration grayscale texture image dataset. According to the evaluation index of the denoising model in step 2), the optimal model parameters are obtained, and the optimal parameters are used to denoise the vibration grayscale texture image datasets with different degrees of noise. After the denoised vibration grayscale texture images are marked, a denoised fault feature dataset is established, and it is divided into a training set and a test set. The training set is used for model training, and the test set verifies the final diagnosis effect, and the final structure and parameters of the fault diagnosis model are saved.

[0046] The specific steps for constructing the mechanical transmission system fault diagnosis model in step 3) are as follows: 3.1) Construct a convolutional neural network fault diagnosis model based on MobileNetV3, including a feature extractor and a classifier, optimize the width of the model structure of the feature extraction part, and simplify the model structure; 3.2) For low-speed working conditions, use step 1) to construct a vibration grayscale texture image dataset of bearings, rotors, and gears in normal and faulty states in the mechanical transmission system, label the denoised images in step 2) to establish a dataset, and divide it into a training set and a test set, which are input into the fault diagnosis model for training and testing. The classification accuracy, precision, recall rate, and F1 score of the fault diagnosis model on the test set are used as evaluation indicators, and the calculation formula is:

[0047]

[0048]

[0049]

[0050] In the formula, ACC, precision, recall and F1 represent the accuracy, precision, recall and F1 score of the test set respectively. TP represents the number of samples that are actually positive and correctly predicted as positive, TN represents the number of samples that are actually negative and correctly predicted as negative; FP represents the number of samples that are actually negative but incorrectly predicted as positive; FN represents the number of samples that are actually positive but incorrectly predicted as negative.

[0051] Embodiment 2 See attached Figure 1 The present invention provides a multi-component intelligent fault diagnosis method for a low-speed mechanical transmission system, which mainly comprises the following steps: 1) A mechanical transmission system fault diagnosis test bench was used to conduct a mechanical transmission system fault simulation experiment. Three bearing fault parts, two gear fault parts, and two rotor fault parts were selected for the experiment, including seven fault states, including bearing inner ring fault, bearing outer ring fault, bearing rolling element fault, gear tooth surface wear, gear tooth root crack, rotor imbalance, and rotor misalignment, as well as one normal state. The experimental working conditions were set at low speed (700r / min) and no load (0N·m). For bearing outer ring, rolling element and gear faults, acceleration sensors were used to collect vibration signals of the test bench shell; for bearing inner ring and rotor faults, eddy current displacement sensors were used to collect vibration signals; 1.1) Collect vibration acceleration and displacement signals of different components and different states to obtain original vibration data, divide the obtained original vibration data into multiple groups with 4096 sampling points as a group without overlap, and splice them row by row to form a feature matrix; 1.2) The elements of the feature matrix are normalized from 0 to 255 and converted into a grayscale image. The conversion relationship between the feature matrix element value and the grayscale image pixel value is:

[0052] In the formula, i , j Represents the position coordinates in the feature matrix or grayscale image, is the feature matrix located at ( i , j ) element, X max and X min Represent the maximum and minimum values ​​in the feature matrix respectively, is the gray value of the corresponding position after transformation; 1.3) Scale the converted grayscale image to 256×256 pixels; 1.4) Perform robust local binarization on the image obtained after equal-proportion scaling to generate a vibration grayscale texture image, and introduce weighted local grayscale values ​​in the robust local binarization process to adjust the weight of the center pixel for threshold selection;

[0053] In the formula, is the gray value of the center pixel, The radius is R The grayscale values ​​of adjacent pixels on the circle, is a constant, The larger the value, the greater the proportion of the center threshold; After completing the threshold selection, robust local binarization is performed to describe the texture features of the image. The size of the circular pixel area in the image is calculated in sequence. The calculation formula for robust local binarization is:

[0054] In the formula, p Represents the total number of selected area points on the circumference, , Represents the gray value of the area point. P,R is the RLBP value of the center point after being processed by the RLBP operator, ; 1.5) Add different degrees of Gaussian noise to all the obtained vibration grayscale texture images to construct a noisy grayscale texture image dataset, and divide it into a training set and a test set according to a preset ratio.

[0055] 2) Construct an autoencoder vibration grayscale texture image denoising model, which specifically includes the following steps: 2.1) Construct an autoencoder vibration grayscale texture image denoising model, including multi-scale convolution, Ghost convolution and ECB convolution; 2.2) Use step 1) to construct a vibration grayscale texture image dataset of bearings, rotors, and gears in mechanical transmission systems under normal and faulty conditions at low speed. Divide the training set and test set into a ratio of 8:2, add three different degrees of Gaussian noise with standard deviations of 15, 25, and 50 to the dataset, and achieve the optimal denoising effect by updating the model hyperparameter settings. The denoising effect of the denoising model is evaluated by the peak signal-to-noise ratio and structural similarity indicators.

[0056] Using five comparison methods and the method of the present invention, the denoising results of the vibration grayscale texture image under different degrees of Gaussian noise are shown in Tables 1 to 3: Table 1 Model denoising results under Gaussian noise with standard deviation of 15

[0057] Table 2 Denoising results of various models under Gaussian noise with standard deviation of 25

[0058] Table 3 Model denoising results under Gaussian noise with standard deviation of 50

[0059] 3) Construct a mechanical transmission system fault diagnosis model, which includes the following steps: 3.1) Construct a convolutional neural network fault diagnosis model, including a feature extractor and a classifier. Optimize the model width of the feature extraction part and simplify the model structure. The designed model structure is shown in the attached figure. Figure 4 As shown; 3.2) For the low-speed and no-load working condition data, the vibration grayscale texture image dataset of the bearings, rotors, and gears in the mechanical transmission system under low-speed and no-load working conditions is constructed using step 1). The vibration grayscale texture image after denoising in step 2) is labeled and then a dataset is established. The dataset is divided into a training set and a test set, which are input into the fault diagnosis model for training and testing. The specific parameter settings of each layer of the final model are shown in Table 4: Table 4 Parameter settings of each layer of the model

[0060] The classification accuracy, precision, recall rate and F1 score of the fault diagnosis model on the test set are used as evaluation indicators. The test accuracy of the final model on each denoising data set is above 96%. Figure 5 The classification results of data sets with different degrees of noise and noise reduction.

[0061] To demonstrate the superiority of the method of the present invention, the method of the present invention is compared with the original MobileNetV3 basic model and two classic convolutional neural networks, GoogLeNet and AlexNet. Also under the same low speed and no-load conditions, the results are shown in Table 5.

[0062] Table 5 Comparison of results of different diagnostic methods

[0063] Embodiment 3 A multi-component intelligent fault diagnosis system for a low-speed mechanical transmission system, comprising: Mechanical transmission system vibration grayscale texture image dataset construction module: collect the vibration acceleration and displacement signals of bearings, rotors, and gears in the mechanical transmission system under normal and faulty conditions at low speed, segment and splice the vibration acceleration and displacement signals into feature matrices according to the specified number of points, and then perform standardization, proportional scaling, and robust local binarization processing in sequence to generate mechanical transmission system vibration grayscale texture images. All the generated images constitute a mechanical transmission system vibration grayscale texture image dataset, and the vibration grayscale texture image dataset is divided into a training set and a test set in proportion; Autoencoder vibration grayscale texture image denoising model construction module: used to use low-speed working condition data to construct a vibration grayscale texture image dataset of low-speed working condition through the operations in the mechanical transmission system vibration grayscale texture image dataset construction module, and use the vibration grayscale texture image dataset of low-speed working condition to construct an autoencoder vibration grayscale texture image denoising model. The autoencoder denoising model includes multi-scale convolution, Ghost convolution and ECB convolution. The training set is used for model training, and the test set is used for model testing. The denoising effect of the denoising model is evaluated by the peak signal-to-noise ratio and structural similarity indicators.

[0064] Mechanical transmission system fault diagnosis model construction module: Construct a convolutional neural network fault diagnosis model, including a feature extractor and a classifier, optimize the width of the model structure of the feature extraction part, and simplify the model structure. For low-speed working condition data, build a vibration grayscale texture image dataset of bearings, rotors, and gears in the mechanical transmission system under normal and fault conditions, label the vibration grayscale texture image samples after denoising by the denoising model to establish a fault feature dataset, and divide it into a training set and a test set to input into the fault diagnosis model for training and testing. The classification accuracy, precision, recall rate, and F1 score of the fault diagnosis model on the test set are used as evaluation indicators to complete the fault diagnosis of the mechanical transmission system under low-speed working conditions and noise interference.

[0065] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0066] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0067] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0068] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit its protection scope. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that after reading the present invention, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the invention, but these changes, modifications or equivalent substitutions are all within the protection scope of the pending claims of the invention.

Claims

1. A multi-component intelligent fault diagnosis method for a low-speed mechanical transmission system, characterized in that: The following steps are involved: 1) Constructing a mechanical transmission system vibration grayscale texture image dataset The vibration acceleration and displacement signals of bearings, rotors and gears in the mechanical transmission system under normal and fault conditions at low speed are collected. The vibration acceleration and displacement signals are segmented and spliced ​​into feature matrices according to the specified number of points, and then standardized, proportionally scaled and robustly locally binarized are processed in sequence to generate the mechanical transmission system vibration grayscale texture image. All the generated images constitute the mechanical transmission system vibration grayscale texture image dataset, and the vibration grayscale texture image dataset is divided into a training set and a test set in proportion. 2) Construct an autoencoder vibration grayscale texture image denoising model Based on the mechanical transmission system vibration grayscale texture image dataset in step 1), a denoising model of the autoencoder vibration grayscale texture image is constructed, wherein the denoising model comprises multi-scale convolution, Ghost convolution and ECB convolution, the training set is used for model training, the test set is used for model testing, and the denoising effect of the denoising model is evaluated by peak signal-to-noise ratio and structural similarity indicators; 3) Construct a fault diagnosis model for mechanical transmission systems Based on the lightweight convolutional neural network, the model width is optimized to construct an improved convolutional neural network fault diagnosis model. Based on the mechanical transmission system vibration grayscale texture image dataset in step 1), a fault feature dataset is established after the fault label is added, and it is divided into a training set and a test set. The training set is used for model training, and the test set verifies the final diagnosis result. The final structure and parameters of the fault diagnosis model are saved. After the model structure and parameters are determined, three different degrees of noise, light, medium and heavy, are added to the mechanical transmission system vibration grayscale texture image dataset in step 1) to establish a noisy vibration grayscale texture image dataset. According to the evaluation index of the denoising model in step 2), the optimal model parameters are obtained, and the optimal parameters are used to denoise the vibration grayscale texture image datasets with different degrees of noise. After the denoised vibration grayscale texture images are labeled, a denoised fault feature dataset is established, and it is divided into a training set and a test set. The training set is used for model training, and the test set verifies the diagnostic effect of the fault diagnosis model on the denoised image dataset.

2. A multi-component intelligent fault diagnosis method for a low-speed mechanical transmission system according to claim 1, characterized in that: The step 1) constructs a mechanical transmission system vibration grayscale texture image dataset, specifically: 1.1) Collect the vibration acceleration and displacement signals of bearings, rotors and gears in the mechanical transmission system under normal and fault conditions at low speed, and segment and splice the vibration acceleration and displacement signals according to the specified number of points to form a feature matrix; 1.2) The elements of the feature matrix are normalized from 0 to 255 and converted into a grayscale image. The conversion relationship between the feature matrix element value and the grayscale image pixel value is: In the formula, i , j Represents the position coordinates in the feature matrix or grayscale image, is the feature matrix located at ( i , j ) element, X max and X min Represent the maximum and minimum values ​​in the feature matrix respectively, is the gray value of the corresponding position after transformation; 1.3) Scale the converted grayscale image to 256×256 pixels; 1.4) Perform robust local binarization on the image obtained after equal-proportion scaling to generate a vibration grayscale texture image, and introduce weighted local grayscale values ​​in the robust local binarization process to adjust the weight of the center pixel for threshold selection; In the formula, is the gray value of the center pixel, The radius is R The grayscale values ​​of adjacent pixels on the circle, is a constant, The larger the value, the greater the proportion of the center threshold; After completing the threshold selection, robust local binarization is performed to describe the texture features of the image. The size of the circular pixel area in the image is calculated in sequence. The calculation formula for robust local binarization is: In the formula, p Represents the total number of selected area points on the circumference, , Represents the gray value of the area point, RLBP P,R is the RLBP value of the center point after being processed by the RLBP operator, ; 1.5) Add different degrees of Gaussian noise to all the obtained vibration grayscale texture images to construct a noisy grayscale texture image dataset, and divide it into a training set and a test set according to a preset ratio.

3. A multi-component intelligent fault diagnosis method for a low-speed mechanical transmission system according to claim 2, characterized in that: The preset ratio is 8:

2.

4. The multi-component intelligent fault diagnosis method for a low-speed mechanical transmission system according to claim 2, characterized in that: The generated image samples contain prominent texture detail features. The autoencoder denoising model is used to perform feature extraction and image reconstruction on the noisy vibration grayscale texture image dataset in step 1.5) to achieve the purpose of denoising.

5. The multi-component intelligent fault diagnosis method for a low-speed mechanical transmission system according to claim 1, characterized in that: In the step 2), an autoencoder vibration grayscale texture image denoising model is constructed, specifically: 2.1) Construct an autoencoder vibration grayscale texture image denoising model, including multi-scale convolution, Ghost convolution and ECB convolution; 2.2) Use step 1) to construct a vibration grayscale texture image dataset of bearings, rotors, and gears in mechanical transmission systems under normal and faulty conditions at low speed. Divide the training set and test set into a ratio of 8:2, add three different degrees of Gaussian noise with standard deviations of 15, 25, and 50 to the dataset, and achieve the optimal denoising effect by updating the model hyperparameter settings. The denoising effect of the denoising model is evaluated by the peak signal-to-noise ratio and structural similarity indicators.

6. The multi-component intelligent fault diagnosis method for a low-speed mechanical transmission system according to claim 1, characterized in that: In step 3), a mechanical transmission system fault diagnosis model is constructed, specifically: 3.1) Construct a convolutional neural network fault diagnosis model, including a feature extractor and a classifier, optimize the width of the model structure of the feature extraction part, and simplify the model structure; 3.2) For low speed working conditions, step 1) is used to construct a vibration grayscale texture image dataset of bearings, rotors, and gears in the mechanical transmission system under normal and faulty conditions. The vibration grayscale texture images after denoising in step 2) are labeled to establish a dataset, which is divided into a training set and a test set. The dataset is input into the fault diagnosis model for training and testing. The classification accuracy, precision, recall rate, and F1 score of the fault diagnosis model on the test set are used as evaluation indicators. The calculation formula is: Where ACC, precision, recall and F1 represent the accuracy, precision, recall and F1 score of the test set respectively; TP represents the number of samples that are actually positive and correctly predicted as positive, TN represents the number of samples that are actually negative and correctly predicted as negative; FP represents the number of samples that are actually negative but incorrectly predicted as positive; FN represents the number of samples that are actually positive but incorrectly predicted as negative.

7. A multi-component intelligent fault diagnosis system for a low-speed mechanical transmission system, characterized in that: include: Mechanical transmission system vibration grayscale texture image dataset construction module: collect the vibration acceleration and displacement signals of bearings, rotors, and gears in the mechanical transmission system under normal and faulty conditions at low speed, segment and splice the vibration acceleration and displacement signals into feature matrices according to the specified number of points, and then perform standardization, proportional scaling, and robust local binarization processing in sequence to generate the corresponding vibration grayscale texture images. All the generated images constitute the mechanical transmission system vibration grayscale texture image dataset, and the vibration grayscale texture image dataset is divided into a training set and a test set in proportion; An autoencoder vibration grayscale texture image denoising model construction module is used to use low-speed working condition data to construct a vibration grayscale texture image dataset of low-speed working condition through operations in a mechanical transmission system vibration grayscale texture image dataset construction module, and to construct an autoencoder vibration grayscale texture image denoising model using the vibration grayscale texture image dataset of low-speed working condition. The autoencoder denoising model includes multi-scale convolution, Ghost convolution and ECB convolution. The training set is used for model training, and the test set is used for model testing. The denoising effect of the denoising model is evaluated by peak signal-to-noise ratio and structural similarity indicators. Mechanical transmission system fault diagnosis model construction module: Construct a convolutional neural network fault diagnosis model, including a feature extractor and a classifier, optimize the width of the model structure of the feature extraction part, simplify the model structure, and build a vibration grayscale texture image dataset of bearings, rotors, and gears in the mechanical transmission system under normal and fault conditions for low-speed working conditions data. After labeling the vibration grayscale texture image samples after denoising by the denoising model, establish a fault feature dataset, and divide it into a training set and a test set to input into the fault diagnosis model for training and testing. The classification accuracy, precision, recall rate and F1 score of the fault diagnosis model on the test set are used as evaluation indicators to complete the fault diagnosis of the mechanical transmission system under low-speed working conditions and noise interference.

8. The multi-component intelligent fault diagnosis system for a low-speed mechanical transmission system according to claim 7, characterized in that: The construction process of the mechanical transmission system vibration grayscale texture image dataset is specifically as follows: 1.1) Collect the vibration acceleration and displacement signals of bearings, rotors and gears in the mechanical transmission system under normal and fault conditions at low speed, and segment and splice the vibration acceleration and displacement signals according to the specified number of points to form a feature matrix; 1.2) The elements of the feature matrix are normalized from 0 to 255 and converted into a grayscale image. The conversion relationship between the feature matrix element value and the grayscale image pixel value is: In the formula, i , j Represents the position coordinates in the feature matrix or grayscale image, is the feature matrix located at ( i , j ) element, X max and X min represents the maximum and minimum values ​​in the feature matrix, respectively. is the gray value of the corresponding position after transformation; 1.3) Scale the converted grayscale image to 256×256 pixels; 1.4) Perform robust local binarization on the image obtained after equal-proportion scaling to generate a vibration grayscale texture image, and introduce weighted local grayscale values ​​in the robust local binarization process to adjust the weight of the center pixel for threshold selection; In the formula, is the gray value of the center pixel, The radius is R The grayscale values ​​of adjacent pixels on the circle, is a constant, The larger the value, the greater the proportion of the center threshold; After completing the threshold selection, robust local binarization is performed to describe the texture features of the image. The size of the circular pixel area in the image is calculated in sequence. The calculation formula for robust local binarization is: In the formula, p Represents the total number of selected area points on the circumference, , Represents the gray value of the area point, RLBP P,R is the RLBP value of the center point after being processed by the RLBP operator, ; 1.5) Add different degrees of Gaussian noise to all the obtained vibration grayscale texture images to construct a noisy grayscale texture image dataset, and divide it into a training set and a test set according to a preset ratio.

9. The multi-component intelligent fault diagnosis system for a low-speed mechanical transmission system according to claim 7, characterized in that: The construction of the autoencoder vibration grayscale texture image denoising model is specifically as follows: 2.1) Construct an autoencoder vibration grayscale texture image denoising model, including multi-scale convolution, Ghost convolution and ECB convolution; 2.2) Use step 1) to construct a vibration grayscale texture image dataset of bearings, rotors, and gears in mechanical transmission systems under normal and faulty conditions at low speed. Divide the training set and test set into a ratio of 8:2, add three different levels of Gaussian noise with standard deviations of 15, 25, and 50 to the dataset, and achieve the optimal denoising effect by updating the model hyperparameter settings. The denoising effect of the denoising model is evaluated by the peak signal-to-noise ratio and structural similarity indicators.

10. The multi-component intelligent fault diagnosis system for a low-speed mechanical transmission system according to claim 7, characterized in that: The construction of the mechanical transmission system fault diagnosis model is specifically as follows: 3.1) Construct a convolutional neural network fault diagnosis model, including a feature extractor and a classifier, optimize the width of the model structure of the feature extraction part, and simplify the model structure; 3.2) For low speed working conditions, step 1) is used to construct a vibration grayscale texture image dataset of bearings, rotors, and gears in the mechanical transmission system under normal and faulty conditions. The vibration grayscale texture images after denoising in step 2) are labeled to establish a dataset, which is divided into a training set and a test set. The dataset is input into the fault diagnosis model for training and testing. The classification accuracy, precision, recall rate, and F1 score of the fault diagnosis model on the test set are used as evaluation indicators. The calculation formula is: Where ACC, precision, recall and F1 represent the accuracy, precision, recall and F1 score of the test set respectively; TP represents the number of samples that are actually positive and correctly predicted as positive, TN represents the number of samples that are actually negative and correctly predicted as negative; FP represents the number of samples that are actually negative but incorrectly predicted as positive; FN represents the number of samples that are actually positive but incorrectly predicted as negative.

Images