Rolling bearing fault identification sensing device based on 24-bit ADC and identification sensing method thereof

Abstract

The invention relates to a rolling bearing fault identification sensing device based on a 24-bit ADC and an identification sensing method thereof, the rolling bearing fault identification sensing device comprises a device body, an accelerometer module is installed in the device body, the output end of the accelerometer module is connected with a signal processing module, the output end of the signal processing module is connected with a data processing device, and the data processing device is connected with an analog-to-digital converter. The output end of the data processing device is connected with a wireless module, the data processing device is provided with an independent power supply control module, and the power supply control module is connected with a power supply assembly. A vibration acceleration signal of the rolling bearing is collected and processed, wherein data segmentation processing and fast Fourier transform processing are included. Therefore, few computing resources are occupied, and long-time operation processing is facilitated. The energy consumption required during the use is low, and the self-energy supply can be realized through the self-contained power supply assembly.

Description

Technical Field

[0001] The present invention relates to a rolling bearing fault detection device based on 24-bit ADC, and in particular to a rolling bearing fault identification sensing device based on 24-bit ADC and an identification sensing method thereof. Background Art

[0002] As one of the essential parts of rotating mechanical equipment, rolling bearings are widely used in my country's industrial production process, and their operating status greatly affects the operating status of rotating mechanical equipment. Because the failure of rolling bearings will cause the entire rotating mechanical equipment to fail. Rotating mechanical equipment will generate vibrations during operation, and the vibration signal will effectively reflect the operating status of the equipment. Specifically, it contains fault information, so collecting and analyzing these vibration signals and extracting relevant information is the most effective and commonly used method for equipment fault diagnosis.

[0003] At present, the more conventional methods in this field are mostly statistical analysis or signal processing methods. In recent years, with the rapid development of artificial intelligence and related fields, fault diagnosis models and methods based on machine learning and deep learning have been widely used.

[0004] However, as the amount of data used in analysis and testing has increased, problems have arisen in which too many computing resources are occupied, power consumption is too high, and diagnostic efficiency cannot be guaranteed.

[0005] In view of the above-mentioned defects, the designers have actively carried out research and innovation in order to create a rolling bearing fault identification sensor device and its identification sensor method to make it more valuable for industrial use. Summary of the invention

[0006] In order to solve the above technical problems, the purpose of the present invention is to provide a rolling bearing fault identification sensing device and an identification sensing method thereof.

[0007] The rolling bearing fault identification sensor device based on 24-bit ADC of the present invention includes a device body, wherein: an accelerometer module is installed in the device body, the output end of the accelerometer module is connected to a signal processing module, the output end of the signal processing module is connected to a data processing device, the output end of the data processing device is connected to a wireless module, the data processing device is configured with an independent power supply control module, the power supply control module is connected to a power supply component, the data processing device includes a data acquisition module connected to the signal processing module, the output end of the data acquisition module is connected to a data preprocessing module, the output end of the data preprocessing module is connected to a deep learning module, the output end of the deep learning module is connected to a fault identification module, the output end of the fault identification module is connected to the wireless module, and the data processing device also includes a data storage module, which is respectively connected to the data storage interfaces of the data acquisition module, the data preprocessing module, the deep learning module, and the fault identification module.

[0008] Furthermore, in the above-mentioned rolling bearing fault identification sensor device based on 24-bit ADC, the accelerometer module is a piezoelectric sensor, the brightness of the piezoelectric sensor is 40mV / g, the frequency response range is 28000Hz, and the acceleration range is 50g, which can collect the early fault signal waveform of the rolling bearing.

[0009] Furthermore, in the above-mentioned rolling bearing fault identification sensor device based on 24-bit ADC, the core of the signal processing module is a 24-bit ADC analog-to-digital conversion chip.

[0010] Furthermore, in the above-mentioned rolling bearing fault identification sensor device based on 24-bit ADC, the wireless module is a module integrating Wi-Fi and Bluetooth.

[0011] Furthermore, in the above-mentioned rolling bearing fault identification sensor device based on 24-bit ADC, the power supply control module is a voltage and current distribution control module; and the power supply component is a combination of a lithium battery and a super capacitor.

[0012] Furthermore, in the above-mentioned rolling bearing fault identification sensor device based on 24-bit ADC, the data acquisition module acquires data from the signal processing module through SPI; and the data preprocessing module is a vibration acceleration signal preprocessing module.

[0013] Furthermore, in the above-mentioned rolling bearing fault identification sensing device based on 24-bit ADC, the deep learning module is a two-dimensional convolutional neural network (2DCNN) model processing module corresponding to the three-axis FFT data in the three-axis vibration acceleration signal data of the rolling bearing.

[0014] Furthermore, in the above-mentioned rolling bearing fault identification sensing device based on 24-bit ADC, the fault identification module is an identification module that performs inference processing on the operating status of the corresponding rolling bearing, and the data storage module is a hexadecimal format storage module, a built-in storage chip, or an external hard disk.

[0015] The rolling bearing fault identification sensing method based on 24-bit ADC is characterized by comprising the following steps:

[0016] Step 1: collecting the vibration acceleration signal of the rolling bearing for processing, including data segmentation processing and fast Fourier transform processing;

[0017] Step 2: Perform model learning on the data processed in step 1;

[0018] Step 3: perform fault identification reasoning;

[0019] Step 4: Send the abnormal data to the wireless terminal.

[0020] Furthermore, the above-mentioned rolling bearing fault identification sensing method based on 24-bit ADC, wherein,.

[0021] In the step 1, the data segmentation includes segmenting and reconstructing a time series signal such as vibration acceleration, and segmenting the rolling bearing vibration signal by a sliding window method to generate a data set;

[0022] Assume that the sliding window length is a, which is usually also the length of the sample, the sliding step length is b, and the total number of sampling points of each type of rolling bearing running state vibration acceleration data is M. The rolling bearing running state vibration acceleration data is obtained, and the calculation formula of the total number of samples N is:

[0023]

[0024] If you want to make the sliding windows non-overlapping, you need to set the sliding step size b equal to the sliding window length a. At this time,

[0025]

[0026] During the actual operation of the rolling bearing, when collecting the rolling bearing vibration acceleration signal data, the rolling bearing speed RS is set to 2500r / min = 41.67r / s, and the sampling frequency Fs of the vibration acceleration sensor is set to 25600Hz.

[0027]

[0028] It can be concluded that for every rotation of the rolling bearing, the vibration sensor collects 614.4 points, and for every 5 rotations of the rolling bearing, the vibration sensor collects 3072 points. The sliding window length, i.e., the sample length, is a=3072, and the data is divided from left to right without overlapping, i.e., b=3072. Each sample contains the acceleration data of the three vibration directions for every 3 rotations of the rolling bearing during continuous rotation.

[0029] The Fourier transform process is to normalize the three axes of the original signal of the vibration acceleration signal of the rolling bearing respectively. The processing formula is as follows:

[0030]

[0031] Perform Fourier transform of the time series with a parameter of 256.

[0032]

[0033] The three-axis FFT signals of the normalized rolling bearing vibration acceleration signal are spliced ​​in the vertical direction to obtain the final fused FFT.

[0034]

[0035] Among them, D x ,D y ,D z They are the FFT of the three-axis vibration acceleration signal of the rolling bearing;

[0036] In the step 2, a neural network layout is performed on the three-axis FFT data of the three-axis vibration acceleration signal data of the rolling bearing.

[0037] The neural network is a two-dimensional convolutional neural network (2DCNN) with 10 layers.

[0038] The three-axis FFT data of the three-axis vibration acceleration signal data of the rolling bearing is used as the input of the network. One layer of convolution and one layer of maximum pooling are used to extract image features. The convolution layer has 32 convolution kernels and outputs 32 feature maps.

[0039] The pooling layer is 2×1 maximum pooling with a step size of 1. A Dropout random deactivation layer is added between the maximum pooling layer and the Flatten layer to randomly return the output or part of the weights of the hidden layer of the network model to zero.

[0040] There are two fully connected layers, and the last fully connected layer is used as the output layer. The output layer has 4 neurons, and 4 is the total number of categories of rolling bearing operating states.

[0041] In the step 3, the loading reasoning process of the model is converted into a set of ai_bfsa_network_XXX() function libraries defined by the AI ​​client, and the model generated by the STM32CubeMX and X-CUBE-AI tools; the ai_bfsa_network_XXX() function library contains initialization, obtaining weight parameters, obtaining activation parameters, input and output functions, and its function references are shown in the following table,

[0042]

[0043]

[0044] After that, through the processing of the data processing device (MCU), the reasoning process is as follows:

[0045] Use ai_bfsa_network_create_and_init() function to create and initialize C-model;

[0046] Use ai_bfsa_network_input_get() and ai_bfsa_network_output_get() functions to define the input and output pointers of the model for storing input and output data;

[0047] Use the AI_HANDLE_PTR() function to update the I / O handler during model inference and execute the inference function ai_bfsa_network_run() to get the output data;

[0048] The error of the running process is obtained from the ai_bfsa_network_get_error() function.

[0049] By means of the above scheme, the present invention has at least the following advantages:

[0050] 1. It takes up very little computing resources, which is convenient for long-term operation and processing.

[0051] 2. The energy consumption required during use is low, and autonomous energy supply can be achieved through its own power supply components.

[0052] 3. The diagnostic efficiency has been improved. After many experiments, the average time taken by the three-axis acceleration vibration sensor to perform a rolling bearing operating status monitoring is 3.199ms.

[0053] 4. Wireless method can be used to achieve real-time data transmission with the background.

[0054] 5. The constructed equipment has a simple structure and is easy to layout and manufacture.

[0055] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present invention in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a structural schematic diagram of a rolling bearing fault identification sensor device based on 24-bit ADC.

[0057] Figure 2 It is a schematic diagram of the two-dimensional convolutional neural network (2DCNN) model.

[0058] Figure 3 It is the FFT data 2DCNN network structure parameters and input and output results.

[0059] Figure 4 , Figure 5 It is the recognition result comparison curve and confusion matrix of the model trained by the method of the present invention.

[0060] Figure 6 This is the model architecture of the two-dimensional convolutional neural network before compression.

[0061] Figure 7 This is the compressed model architecture of the two-dimensional convolutional neural network.

[0062] Figure 8 It is a schematic diagram of the fault identification reasoning process.

[0063] Fig. 9 It is a comparison chart (including matrix) for the verification of the accuracy and real-time performance of rolling bearing fault identification sensors.

[0064] Fig.10 It is a schematic diagram comparing the time taken to process the market using the method of the present invention with that taken using the existing method.

[0065] The meanings of the reference numerals in the drawings are as follows.

[0066] 1 Device body 2 Accelerometer module

[0067] 3 Signal processing module 4 Data processing device

[0068] 5 Wireless module 6 Power supply control module

[0069] 7 Power supply assembly 8 Data acquisition module

[0070] 9 Data preprocessing module 10 Deep learning module

[0071] 11 Fault identification module 12 Data storage module

[0072] 13 Data acquisition port DETAILED DESCRIPTION

[0073] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0074] like Figure 1 The rolling bearing fault identification sensor device based on 24-bit ADC includes a device body 1, which is different from the others in that an accelerometer module 2 is installed in the device body 1. The accelerometer module 2 is provided with a data acquisition port 12 connected to the rolling bearing for collecting vibration signals. The output end of the accelerometer module 2 is connected with a signal processing module 3 for realizing analog-to-digital conversion. At the same time, the output end of the signal processing module 3 is connected with a data processing device 4, which can process the acquired data to determine whether there is a fault. In addition, considering the ability to communicate data with a wireless terminal in the background, the output end of the data processing device 4 is connected with a wireless module 5. Furthermore, in order to ensure stable operation between the modules, the data processing device 4 is provided with an independent power supply control module 6, and a power supply component 7 is connected to the power supply control module 6.

[0075] In order to facilitate the implementation of the present invention, the accelerometer module used is commercially available model 540C-50. The signal processing module is optionally available in model MA24873, and can preferably be a module of the same level prepared by the applicant. At the same time, the data processing device used can be optionally equipped with an MCU of model STM32L471RGT6. In addition, the power control module used can be selected from TPS780270200DDC and NCP718ASN300T1G models. The power supply component is an ER26500+SPC1550 type battery. The wireless module is a FCE863R Wi-Fi and Bluetooth integrated module.

[0076] During the implementation, the data processing device 4 includes a data acquisition module 8 connected to the signal processing module 3 to facilitate data acquisition. The output end of the data acquisition module 8 is connected to the data preprocessing module 9, and the output end of the data preprocessing module 9 is connected to the deep learning module 10. At the same time, the output end of the deep learning module 10 is connected to the fault identification module 11. In this way, the recognition accuracy can be improved through data training. Considering the convenience of data transmission, the output end of the fault identification module 11 is connected to the wireless module 5. Furthermore, the data processing device 4 also includes a data storage module 12. The data storage module 12 is respectively connected to the data storage interfaces of the data acquisition module 8, the data preprocessing module 9, the deep learning module 10, and the fault identification module 11. As a result, it is convenient to store various intermediate values ​​and final data.

[0077] In combination with a preferred embodiment of the present invention, the accelerometer module 2 is a piezoelectric sensor, the lumen of the piezoelectric sensor is 40mV / g, its frequency response range is 28000Hz, and the acceleration range is 50g, which can collect the very early fault signal waveform of the rolling bearing. In this way, the abnormality can be detected in time at the initial stage, which is convenient for the background to intervene and handle in the early stage.

[0078] Further, the signal processing module 3 used in the present invention is a 24-bit ADC analog-to-digital conversion chip. Specifically, 24 bits represent 24 bits, i.e., 2 24 =16777216, the voltage of the accelerometer full scale 50g corresponds to 5V, that is, the resolution of 24bitADC is 5V / 16777216=0.000000298V, that is, 298nV, corresponding to an acceleration of 2.98ug. At the same time, considering the stability of data transmission and having better wireless data compatibility, the wireless module 5 is a module integrating Wi-Fi and Bluetooth. In this way, the fault identification result of the rolling bearing can be uploaded to the wireless terminal via Wi-Fi or Bluetooth.

[0079] In view of the actual implementation, the power supply control module 6 is a voltage and current distribution control module, which can meet the voltage and current required for the operation of each component. Of course, considering the need for continuous energy supply, the power supply component 7 used is a lithium battery. At the same time, the data acquisition module 8 used is a hexadecimal format storage module. The data preprocessing module 9 is a vibration acceleration signal preprocessing module. Thus, the original vibration acceleration signal data can be processed to obtain a frequency domain signal.

[0080] Looking further, the deep learning module 10 used is a two-dimensional convolutional neural network (2DCNN) model processing module corresponding to the three-axis FFT data in the three-axis vibration acceleration signal data of the rolling bearing. In other words, the three-axis FFT data of the three-axis vibration acceleration signal data of the rolling bearing can be trained using a two-dimensional convolutional neural network (2DCNN). At the same time, the fault identification module 11 is an identification module that performs reasoning processing on the operating state of the corresponding rolling bearing. Thus, the acceleration frequency domain data processed by the data preprocessing module 9 can be loaded into the deep learning module 10. Furthermore, the data storage module 12 is a built-in storage chip or an external hard disk.

[0081] In order to better implement the present invention, a rolling bearing fault identification sensing method is now provided, which comprises the following steps:

[0082] Step 1: collect the vibration acceleration signal of the rolling bearing for processing, including data segmentation processing and fast Fourier transform processing.

[0083] Specifically, the data segmentation adopted is to segment and reconstruct the time series signals such as vibration acceleration, and generate data sets by segmenting the rolling bearing vibration signals through the sliding window method;

[0084] Assume that the sliding window length is a, which is usually also the length of the sample, the sliding step length is b, and the total number of sampling points of each type of rolling bearing running state vibration acceleration data is M. The rolling bearing running state vibration acceleration data is obtained, and the calculation formula of the total number of samples N is:

[0085]

[0086] If you want to make the sliding windows non-overlapping, you need to set the sliding step size b equal to the sliding window length a. At this time,

[0087]

[0088] During the actual operation of the rolling bearing, when collecting the rolling bearing vibration acceleration signal data, the rolling bearing speed RS is set to 2500r / min = 41.67r / s, and the sampling frequency Fs of the vibration acceleration sensor is set to 25600Hz.

[0089]

[0090] It can be concluded that for every rotation of the rolling bearing, the vibration sensor collects 614.4 points, and for every 5 rotations of the rolling bearing, the vibration sensor collects 3072 points.

[0091] At the same time, the sliding window length, i.e., the sample length a=3072, is taken, and the data is divided from left to right without overlapping, i.e., b=3072. Each sample contains acceleration data in three vibration directions during every three rotations of the rolling bearing during continuous rotation.

[0092] In addition, the Fourier transform processing used is to normalize the three axes of the original signal of the vibration acceleration signal of the rolling bearing respectively. The processing formula is as follows:

[0093]

[0094] The data division rules involved are the same as the above operation, and the length of each sample is 3072 points. The time series is subjected to Fourier transform with a parameter of 256.

[0095]

[0096] Due to the symmetry of Fourier transform, the FFT data is halved. The three-axis FFT signals of the normalized rolling bearing vibration acceleration signal are spliced ​​in the vertical direction to obtain the final fused FFT.

[0097]

[0098] Among them, D x ,D y ,D z They are the FFT of the three-axis vibration acceleration signals of the rolling bearing.

[0099] Step 2, the data processed in step 1 is subjected to model learning processing. During this period, the three-axis FFT data of the three-axis vibration acceleration signal data of the rolling bearing is laid out in a neural network. Specifically, the neural network is a two-dimensional convolutional neural network (2DCNN) with a total of 10 layers. The three-axis FFT data of the three-axis vibration acceleration signal data of the rolling bearing is used as the input of the network. In order to reduce the volume of the model, one layer of convolution and one layer of maximum pooling are used to extract image features. The convolution layer has a total of 32 convolution kernels and outputs 32 feature maps. At the same time, the pooling layer is 2×1 maximum pooling with a step size of 1. In this way, a Dropout random deactivation layer is added between the maximum pooling layer and the Flatten layer to randomly return the output or part of the weight of the hidden layer of the network model to zero. In this way, the mutual dependence between nodes can be reduced, thereby realizing the regularization of the two-dimensional convolutional neural network, reducing its structural risk, and alleviating the overfitting of the network. In addition, two layers of fully connected layers can be provided, and the last layer of fully connected layers is used as the output layer, and the output layer has a total of 4 neurons.

[0100] Specifically, the two-dimensional convolutional neural network (2DCNN) is used to train the three-axis FFT data of the three-axis vibration acceleration signal data of the rolling bearing. The two-dimensional convolutional neural network (2DCNN) model designed for the three-axis FFT data of the three-axis vibration acceleration signal data of the rolling bearing is as follows: Figure 2 shown.

[0101] After that, set the FFT data 2DCNN network structure parameters and input and output results, and obtain the analysis result parameters, such as Figure 3 As shown in the table.

[0102] The model provided by the present invention is used for training, and the recognition result comparison curve and confusion matrix of the model are shown in FIG. Figure 4 , Figure 5 shown.

[0103] In addition, during the implementation of the present invention, relying on STM32CubeMX and X-CUBE-AI expansion packages, Figure 2 The 2D CNN model shown in Figure 1 is compressed. During implementation, the model compression does not compress the fully connected layer to ensure the diagnostic accuracy. The intermediate layers are merged and the R / W blocks are defined. Figure 6 This is the model framework before compression. Figure 7The compressed model framework.

[0104] In this way, compared Figure 2 From the above, the flatten layer of the original model is transposed and reshaped, and together with the fully connected fc layer, forms the MatMul layer. At the same time, after the model is compressed, the dropout layer of the original model is deleted. The convolution layer is subdivided into the accumulation of the convolution calculation layer (Conv2D) and the nonlinear calculation layer (Conv2D_Nonlinearity). At the same time, the fully connected layer is subdivided into the accumulation of the fully connected calculation layer (Dense) and the nonlinear calculation layer (Dense_Nonlinearity).

[0105] During implementation, the data processing device (MCU) used in the present invention can be integrated into a 32-bit operating system of the STM32L471RGT6 model. Its main frequency is as high as 168MHz, and the average time for reasoning 1000 times in the MCU is 0.228ms, which has very high real-time performance.

[0106] The following table shows the layers of the compressed model architecture, the time spent in the inference process, and the proportion of the total time. Convolution takes the highest proportion of the time in the inference process.

[0107]

[0108]

[0109] Step 3: Fault identification reasoning. During this period, the acceleration frequency domain data processed by the data preprocessing module is loaded into the deep learning module to reason about the running state of the rolling bearing. The reasoning process is as follows: Figure 8 As shown. AIClient AppLayer represents the AI ​​client application layer; Activation Buffers represents the activation buffer area, which is used to store intermediate variables. Execution file (bfsa_network.c / .h) represents the neural network execution function for rolling bearing fault identification. Data file (bfsa_network_data.c / .h) represents the neural network data file for rolling bearing fault identification. Network runtimeStatic library represents the neural network function standard library. ArmCMSISLibrary represents the Arm function standard library. MCUArmTool-chain represents the tool chain for Arm type MCU.

[0110] In order to better implement the present invention, the loading and reasoning process of the model can be converted into a set of ai_bfsa_network_XXX() function libraries defined by the AI ​​client, and the model generated by the STM32CubeMX and X-CUBE-AI tools; the ai_bfsa_network_XXX() function library contains initialization, obtaining weight parameters, obtaining activation parameters, input and output functions, and its function refers to the following table:

[0111]

[0112]

[0113] After that, through the processing of the data processing device (MCU), the reasoning process is as follows:

[0114] Use the ai_bfsa_network_create_and_init() function to create and initialize the C-model.

[0115] Use ai_bfsa_network_input_get() function and ai_bfsa_network_output_get() function to define the input and output pointers of the model for storing input and output data.

[0116] Use the AI_HANDLE_PTR() function to update the I / O handler during the model inference process and execute the inference function ai_bfsa_network_run() to get the output data.

[0117] The error of the running process is obtained from the ai_bfsa_network_get_error() function.

[0118] Step 4: Send the abnormal data to the wireless terminal.

[0119] The recognition accuracy and real-time performance of the rolling bearing fault recognition sensor device provided by the present invention are verified. Through serial communication, the neural network transplanted into the MCU recognizes and verifies 1000 samples of the test set and returns the recognition results. The comparison curve and confusion matrix of the recognition results are shown in Figure 2. Fig. 9 As shown. Among them, a represents the normal operation of the rolling bearing, b represents the inner ring failure of the rolling bearing, c represents the outer ring failure of the rolling bearing, and d represents the rolling element failure of the rolling bearing. Fig. 9 It can be seen that the MCU's recognition accuracy for the rolling bearing's operating status reaches 99.8%.

[0120] Take the verification of rolling bearing fault identification sensor recognition accuracy and real-time performance as an example:

[0121] The C program based on Keil MDK-ARM is generated by STM32CubeMX and loaded into the MCU (STM32L471RGT6) of the three-axis acceleration vibration sensor through ST-Link. The real-time monitoring process of the rolling bearing operating status is divided into three functions: data acquisition and preprocessing data_acquire_and_preprocess(), model inference aiRun() and inference conclusion Inference_conclusion().

[0122] The running time of the three functions is obtained by running the three functions in a single step. After many experiments, the average time taken by the three-axis acceleration vibration sensor to perform a rolling bearing running status monitoring is 3.199ms.

[0123] The overall processing time after adopting the present invention is compared with the existing method. Fig.10 As shown in the figure, in a diagnosis process, the data collection and preprocessing time is the longest, averaging 2.432ms, followed by the reasoning conclusion module, averaging 0.532ms. The model provided by the present invention takes the least time to reason, with an average measured time of 0.235ms, which is very close to the reasoning time of 0.228ms obtained in the aforementioned hardware verification process.

[0124] It can be seen from the above textual description and the accompanying drawings that the present invention has the following advantages:

[0125] 1. It takes up very little computing resources, which is convenient for long-term operation and processing.

[0126] 2. The energy consumption required during use is low, and autonomous energy supply can be achieved through its own power supply components.

[0127] 3. The diagnostic efficiency has been improved. After many experiments, the average time taken by the three-axis acceleration vibration sensor to perform a rolling bearing operating status monitoring is 3.199ms.

[0128] 4. Wireless method can be used to achieve real-time data transmission with the background.

[0129] 5. The constructed equipment has a simple structure and is easy to layout and manufacture.

[0130] In addition, the indicated orientations or positional relationships described in the present invention are all based on the orientations or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present invention and simplifying the description. They do not indicate or imply that the device or structure referred to must have a specific orientation or be operated with a specific orientation structure. Therefore, they cannot be understood as limitations on the present invention.

[0131] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the technical principles of the present invention, and these improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A rolling bearing fault identification sensor device based on 24-bit ADC includes a device body, characterized in that: An accelerometer module is installed in the device body, the output end of the accelerometer module is connected to a signal processing module, the output end of the signal processing module is connected to a data processing device, the output end of the data processing device is connected to a wireless module, the data processing device is configured with an independent power supply control module, the power supply control module is connected to a power supply component, the data processing device includes a data acquisition module connected to the signal processing module, the output end of the data acquisition module is connected to a data preprocessing module, the output end of the data preprocessing module is connected to a deep learning module, the output end of the deep learning module is connected to a fault identification module, the output end of the fault identification module is connected to the wireless module, the data processing device also includes a data storage module, the data storage module is respectively connected to the data storage interfaces of the data acquisition module, the data preprocessing module, the deep learning module, and the fault identification module.

2. The rolling bearing fault identification sensor device based on 24-bit ADC according to claim 1 is characterized in that: The accelerometer module is a piezoelectric sensor with a sensitivity of 40mV / g, a frequency response range of 28000Hz, and an acceleration range of 50g, and can collect early fault signal waveforms of rolling bearings.

3. The rolling bearing fault identification sensor device based on 24-bit ADC according to claim 1 is characterized in that: The core of the signal processing module is a 24-bit ADC analog-to-digital conversion chip.

4. The rolling bearing fault identification sensor device according to claim 1, characterized in that: The wireless module is a module integrating Wi-Fi and Bluetooth.

5. The rolling bearing fault identification sensor device based on 24-bit ADC according to claim 1 is characterized in that: The power supply control module is a voltage and current distribution control module; the power supply component is a combination of a lithium battery and a super capacitor.

6. The rolling bearing fault identification sensor device based on 24-bit ADC according to claim 1 is characterized in that: The data acquisition module acquires data from the signal processing module through SPI; the data preprocessing module is a vibration acceleration signal preprocessing module.

7. The rolling bearing fault identification sensor device based on 24-bit ADC according to claim 1 is characterized in that: The deep learning module is a two-dimensional convolutional neural network model processing module corresponding to the three-axis FFT data in the three-axis vibration acceleration signal data of the rolling bearing.

8. The rolling bearing fault identification sensor device based on 24-bit ADC according to claim 1 is characterized in that: The fault identification module is an identification module that performs reasoning processing on the operating state of the corresponding rolling bearing, and the data storage module is a hexadecimal format storage module with a built-in storage chip or an external hard disk.

9. A rolling bearing fault identification sensing method based on 24-bit ADC is characterized in that The following steps are involved: Step 1: collecting the vibration acceleration signal of the rolling bearing for processing, including data segmentation processing and fast Fourier transform processing; Step 2: Perform model learning on the data processed in step 1; Step 3: perform fault identification reasoning; Step 4: Send the abnormal data to the wireless terminal.

10. The rolling bearing fault identification sensing method based on 24-bit ADC according to claim 9 is characterized in that: In the step 1, the data segmentation includes segmenting and reconstructing a time series signal such as vibration acceleration, and segmenting the rolling bearing vibration signal by a sliding window method to generate a data set; Assume that the sliding window length is a, which is usually also the length of the sample, the sliding step length is b, and the total number of sampling points of each type of rolling bearing running state vibration acceleration data is M. The rolling bearing running state vibration acceleration data is obtained, and the calculation formula for the total number of samples N is: If you want to make the sliding windows non-overlapping, you need to set the sliding step size b equal to the sliding window length a. At this time, During the actual operation of the rolling bearing, when collecting the rolling bearing vibration acceleration signal data, the rolling bearing speed RS is set to 2500r / min = 41.67r / s, and the sampling frequency Fs of the vibration acceleration sensor is set to 25600Hz. It can be concluded that for every rotation of the rolling bearing, the vibration sensor collects 614.4 points, and for every 5 rotations of the rolling bearing, the vibration sensor collects 3072 points. The sliding window length, i.e., the sample length, is a=3072, and the data is divided from left to right without overlapping, i.e., b=3072. Each sample contains the acceleration data of the three vibration directions for every 3 rotations of the rolling bearing during continuous rotation. The Fourier transform process is to normalize the three axes of the original signal of the vibration acceleration signal of the rolling bearing respectively. The processing formula is as follows: Perform Fourier transform of the time series with a parameter of 256. The three-axis FFT signals of the normalized rolling bearing vibration acceleration signal are spliced ​​in the vertical direction to obtain the final fused FFT. Among them, D x ,D y ,D z They are the FFT of the three-axis vibration acceleration signal of the rolling bearing; In the step 2, a neural network layout is performed on the three-axis FFT data of the three-axis vibration acceleration signal data of the rolling bearing. The neural network is a two-dimensional convolutional neural network (2DCNN) with 10 layers. The three-axis FFT data of the three-axis vibration acceleration signal data of the rolling bearing is used as the input of the network. One layer of convolution and one layer of maximum pooling are used to extract image features. The convolution layer has 32 convolution kernels and outputs 32 feature maps. The pooling layer is 2×1 maximum pooling with a step size of 1. A Dropout random deactivation layer is added between the maximum pooling layer and the Flatten layer to randomly return the output or part of the weights of the hidden layer of the network model to zero. There are two fully connected layers, and the last fully connected layer is used as the output layer. The output layer has 4 neurons, and 4 is the total number of categories of rolling bearing operating states. In the step 3, the loading reasoning process of the model is converted into a set of ai_bfsa_network_XXX() function libraries defined by the AI ​​client, and the model generated by the STM32CubeMX and X-CUBE-AI tools; the ai_bfsa_network_XXX() function library contains initialization, obtaining weight parameters, obtaining activation parameters, input and output functions, and its function references are shown in the following table, After that, through the processing of the data processing device (MCU), the reasoning process is as follows: Use ai_bfsa_network_create_and_init() function to create and initialize C-model; Use ai_bfsa_network_input_get() and ai_bfsa_network_output_get() functions to define the input and output pointers of the model for storing input and output data; Use the AI_HANDLE_PTR() function to update the I / O handler during model inference and execute the inference function ai_bfsa_network_run() to get the output data; The error of the running process is obtained from the ai_bfsa_network_get_error() function.

Images