Bearing fault diagnosis method, system, equipment and medium based on multi-channel analog filtering characteristic network

Abstract

The present invention belongs to the technical field of mechanical fault diagnosis and signal processing, and specifically relates to a bearing fault diagnosis method, system, equipment and medium based on a multi-channel analog filter feature network. The method constructs a multi-channel analog filter feature extraction network, uses multiple parallel bandpass filter channels to extract bearing vibration signal characteristics, and combines an analog neural network classifier to perform fault classification. Signal preprocessing includes removing DC components, amplitude normalization and segmented processing; the filter passband parameters and classifier parameters are jointly optimized through a particle swarm algorithm, with classification accuracy and decision confidence as fitness functions. The system includes signal acquisition, preprocessing, multi-channel filtering, feature extraction, analog neural network classification and result display modules, achieving low-power, high-real-time bearing fault diagnosis. The present invention solves the problems of high power consumption and large delay of traditional digital processing methods, and is suitable for long-term monitoring and large-scale deployment in industrial sites.

Description

Technical Field

[0001] The present invention belongs to the technical field of mechanical fault diagnosis and signal processing, and in particular relates to a bearing fault diagnosis method, system, equipment and medium based on a multi-channel analog filtering characteristic network. Background Art

[0002] Existing bearing fault diagnosis technologies primarily utilize a combination of digital signal processing and machine learning. Traditional approaches typically acquire vibration signals through a high-speed ADC, then employ digital filters to extract frequency-domain features. Classification and diagnosis are then performed using algorithms such as support vector machines, random forests, or deep learning. In recent years, some studies have attempted to deploy neural network models on embedded platforms, such as implementing lightweight CNN network inference based on microcontrollers such as the STM32. While these approaches improve diagnostic accuracy, they rely on high-performance digital processing chips to perform complex feature extraction and model calculations, require high-speed ADCs for signal acquisition, and consume significant computing resources. Furthermore, some studies have employed digital filter banks in conjunction with feature extraction algorithms. While this approach reduces computational complexity, it still requires a complete digital signal processing chain, including anti-aliasing filtering, high-speed sampling, and digital filtering.

[0003] Existing technologies suffer from three major drawbacks: First, digital processing solutions require high-speed ADCs and powerful computing power, resulting in high system power consumption and making it difficult to meet the low-power requirements of long-term industrial monitoring. Second, the sampling and quantization steps introduced by the digital signal processing chain introduce additional latency, affecting the real-time nature of diagnostics. Finally, existing embedded deployment solutions typically require a complete digital computing platform, resulting in high system complexity and high cost, making them unsuitable for large-scale application. While analog circuits offer advantages in power consumption and real-time performance, existing analog filtering solutions often use fixed-parameter filters, which cannot adaptively optimize the feature extraction frequency band and lack collaborative optimization with subsequent classification algorithms, resulting in limited diagnostic performance. Summary of the Invention

[0004] The purpose of the present invention is to provide a bearing fault diagnosis method, system, equipment and medium based on a multi-channel analog filtering feature network to solve the technical problems of high power consumption, poor real-time performance and difficulty in edge deployment caused by traditional bearing fault diagnosis methods due to reliance on digital signal processing.

[0005] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0006] In a first aspect, the present invention proposes a bearing fault diagnosis method based on a multi-channel analog filtering characteristic network, the method comprising:

[0007] Constructing a multi-channel analog filtering feature extraction network, the network comprising a plurality of bandpass filter channels arranged in parallel, each channel being configured with different passband parameters, the passband parameters including a lower cutoff frequency and an upper cutoff frequency;

[0008] Collect bearing vibration signals and perform signal preprocessing;

[0009] Inputting the preprocessed signal into the multi-channel analog filtering feature extraction network, and calculating the root mean square value of the output signal of each channel as the feature value;

[0010] Inputting the characteristic value into a simulated neural network classifier for fault classification;

[0011] The passband parameters of the bandpass filter channel and the parameters of the simulated neural network classifier are jointly optimized in combination with the classification accuracy and decision confidence.

[0012] Preferably, the signal preprocessing includes:

[0013] The DC component in the original vibration signal is removed by a high-pass filter to obtain an AC vibration signal;

[0014] performing amplitude normalization processing on the AC vibration signal to obtain a standardized vibration signal;

[0015] The standardized vibration signal is divided into fixed time windows to obtain signal segments to be analyzed.

[0016] Preferably, the multi-channel analog filtering feature extraction network includes four independent fourth-order Butterworth bandpass filter channels, and the upper and lower cutoff frequencies of each channel are determined by the following method:

[0017] Initialize the cutoff frequency range of each channel based on the characteristic frequency distribution range of the bearing fault;

[0018] The particle swarm optimization algorithm is used to iteratively optimize the upper and lower cutoff frequencies of each channel, with the weighted sum of classification accuracy and decision confidence as the fitness function;

[0019] Save the optimal cutoff frequency combination that maximizes the fitness function.

[0020] Preferably, the working process of the simulated neural network classifier includes:

[0021] Inputting the four-dimensional eigenvalues ​​and corresponding weight parameters into the analog multiplier array according to a predetermined timing sequence through a timing control circuit;

[0022] The multiplication result is input into the integration circuit for weighted summation operation;

[0023] The steady-state voltage values ​​of the four output neurons are compared, and the fault type corresponding to the neuron with the largest voltage value is output as the diagnosis result.

[0024] Preferably, the fitness function used in the joint optimization process is:

[0025] F = α·N(Accuracy) + β·N(Confidence)

[0026] Where N(·) represents normalization processing, α and β are adjustable weight coefficients, and α+β=1;

[0027] The decision confidence is defined as:

[0028] Confidence = max(y_i) - second_max(y_i)

[0029] Where y_i represents the output value of the i-th neuron, and second_max(y_i) represents the second largest output value.

[0030] In a second aspect, the present invention proposes a bearing fault diagnosis system based on a multi-channel analog filtering characteristic network, which is applied to perform the above-mentioned bearing fault diagnosis method. The system includes:

[0031] Signal acquisition module, used to obtain original bearing vibration signals;

[0032] Signal preprocessing module, used to denoise and standardize the original signal;

[0033] Multi-channel filtering module, including multiple bandpass filters set in parallel;

[0034] Feature extraction module, used to calculate the characteristic value of the output signal of each channel;

[0035] Simulated neural network classification module for fault classification based on feature values;

[0036] Result display module, used for outputting diagnosis results;

[0037] The parameters of the multi-channel filtering module and the simulated neural network classification module are pre-configured by an offline optimization module;

[0038] The simulated neural network classification module includes:

[0039] A timing control unit, configured to input the four-dimensional eigenvalues ​​and corresponding weight parameters into the analog multiplier array according to a predetermined timing sequence through a timing control circuit;

[0040] An analog multiplication-addition unit, comprising an analog multiplier array and an integration circuit, inputs the multiplication result into the integration circuit for weighted summation operation;

[0041] The comparison output unit is used to compare the steady-state voltage values ​​of the four output neurons and output the fault type corresponding to the neuron with the largest voltage value as the diagnosis result.

[0042] Preferably, the feature extraction module uses a fourth-order Butterworth bandpass filter and a root mean square detection circuit to extract four-dimensional features; the result display module is used to compare the output voltages of four neurons, encode the channel corresponding to the maximum value as 1, and the rest as 0, and output the fault type.

[0043] Preferably, the system further includes a parameter setting module, which includes:

[0044] Parameter input interface, used to receive filter cutoff frequency and neural network weight parameters;

[0045] A digital-to-analog conversion unit converts digital parameters into corresponding analog voltage signals;

[0046] The parameter distribution unit outputs the converted analog signals to:

[0047] Each bandpass filter of the multi-channel filtering module is configured with its passband parameters;

[0048] The multiplier array of the simulated neural network classification module provides weight parameters;

[0049] Among them, the working timing of the parameter setting module is synchronized with the timing control unit of the simulated neural network classification module.

[0050] In a third aspect, the present invention proposes an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned bearing fault diagnosis method when executing the computer program.

[0051] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, which implements the steps of the bearing fault diagnosis method described above when executed by a processor.

[0052] The beneficial effects of the present invention are:

[0053] The present invention achieves significant optimization of the bearing fault diagnosis system by constructing a collaborative architecture of a multi-channel analog filter feature network and an analog neural network classifier. Compared with traditional digital processing methods, this solution has three core advantages: first, the use of a pure analog signal processing path avoids the resource consumption of high-speed analog-to-digital conversion and digital calculation, greatly reducing system power consumption; second, the particle swarm algorithm is used to jointly optimize the filter passband and neural network parameters, ensuring the collaborative adaptability of feature extraction and classification, and improving diagnostic accuracy; finally, the innovative timing control analog computing architecture achieves real-time response in microseconds. These technical features enable the system to have low power consumption, high real-time performance and strong adaptability, effectively solving the key problems in long-term monitoring and large-scale deployment in industrial sites, and providing a reliable technical solution for rotating machinery status monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A flow chart of a bearing fault diagnosis method based on a multi-channel analog filtering characteristic network proposed in the present invention;

[0055] Figure 2 A schematic diagram of a process for parameter optimization implementation in a specific embodiment of the present invention;

[0056] Figure 3 This is a schematic diagram of the overall hardware implementation of the reasoning process in a specific embodiment of the present invention;

[0057] Figure 4 A schematic diagram of a flow chart of hardware implementation of a parameter setting module in a specific embodiment of the present invention;

[0058] Figure 5 A schematic diagram of a flow chart of hardware implementation of a feature extraction module in a specific embodiment of the present invention;

[0059] Figure 6 A flowchart for the hardware implementation of the digital reasoning unit. DETAILED DESCRIPTION

[0060] The following description provides specific application scenarios and requirements for this specification, with the goal of enabling those skilled in the art to make and use the contents of this specification. Various modifications to the disclosed embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other embodiments and applications without departing from the spirit and scope of this specification. Therefore, this specification is not limited to the embodiments shown, but is intended to be accorded the broadest scope consistent with the claims.

[0061] The terms used herein are for the purpose of describing specific example embodiments only and are not intended to be limiting. For example, as used herein, the singular forms "a," "an," and "the" may also include the plural forms unless the context clearly indicates otherwise. When used in this specification, the terms "comprise," "include," and / or "contain" are intended to refer to the presence of the associated integers, steps, operations, elements, and / or components, but do not preclude the presence of one or more other features, integers, steps, operations, elements, components, and / or groups or the addition of other features, integers, steps, operations, elements, components, and / or groups in the system / method.

[0062] These and other features of this specification, as well as the operation and function of the associated elements of the structure, and the economical assembly and manufacture of the components, can be significantly improved with consideration of the following description. Reference is made to the accompanying drawings, all of which form a part of this specification. However, it should be expressly understood that the drawings are for illustration and description purposes only and are not intended to limit the scope of this specification. It should also be understood that the drawings are not drawn to scale.

[0063] The flowcharts used in this specification illustrate operations implemented by systems according to some embodiments of the present specification. It should be clearly understood that the operations of the flowcharts may not be implemented in sequence. Rather, the operations may be implemented in reverse order or simultaneously. Furthermore, one or more additional operations may be added to the flowcharts. One or more operations may be removed from the flowcharts.

[0064] Example 1

[0065] like Figure 1 As shown, an embodiment of the present application proposes a bearing fault diagnosis method based on a multi-channel analog filtering feature network, the method comprising: constructing a multi-channel analog filtering feature extraction network, the network comprising multiple bandpass filter channels arranged in parallel, each channel configured with different passband parameters, the passband parameters comprising a lower cutoff frequency and an upper cutoff frequency; collecting bearing vibration signals and performing signal preprocessing; inputting the preprocessed signals into the multi-channel analog filtering feature extraction network, and calculating the root mean square value of the output signal of each channel as a eigenvalue; inputting the eigenvalues ​​into an analog neural network classifier for fault classification; wherein, the passband parameters of the bandpass filter channel and the parameters of the analog neural network classifier are jointly optimized in combination with the classification accuracy and decision confidence.

[0066] According to the above embodiment, the present application realizes efficient fault diagnosis by constructing an analog feature extraction network containing multiple parallel bandpass filter channels and combining it with an analog neural network classifier. In the specific implementation, the bearing vibration signal is first subjected to DC component removal, amplitude normalization and segmented preprocessing, and then the signal is input into a multi-channel network composed of a fourth-order Butterworth bandpass filter, and the root mean square value of the output signal of each channel is extracted as a feature vector; the particle swarm algorithm is used to jointly optimize the filter passband parameters and the neural network weights, and the weighted sum of the classification accuracy and decision confidence is used as the fitness function to ensure the coordinated optimization of feature extraction and classification.

[0067] In an optional embodiment, signal preprocessing includes: removing the DC component in the original vibration signal through a high-pass filter to obtain an AC vibration signal; performing amplitude normalization processing on the AC vibration signal to obtain a standardized vibration signal; subtracting the shunt component from the data and then performing a normalization operation to eliminate the constant offset, highlight only the effective signal part, and scale the signal size to a specific range to avoid affecting the feature extraction work; dividing the standardized vibration signal according to a fixed time window to obtain signal segments to be analyzed.

[0068] In an optional embodiment, the multi-channel analog filtering feature extraction network includes four independent fourth-order Butterworth bandpass filter channels, and the upper and lower cutoff frequencies of each channel are determined in the following manner: based on the characteristic frequency distribution range of the bearing fault, the cutoff frequency range of each channel is initialized; using the particle swarm optimization algorithm, the weighted sum of the classification accuracy and the decision confidence is used as the fitness function, and the upper and lower cutoff frequencies of each channel are iteratively optimized; and the optimal cutoff frequency combination that maximizes the fitness function is saved.

[0069] In a preferred embodiment of the present invention, the multi-channel analog filtering feature extraction network utilizes four independent fourth-order Butterworth bandpass filter channels. The passband parameters of each channel are dynamically optimized and configured in the following manner: First, based on the characteristic frequency distribution range of typical bearing faults (such as inner race, outer race, and ball bearing faults), the fundamental frequency and harmonic components of each fault are calculated in combination with bearing geometry and speed, and the cutoff frequency ranges of the four channels are preliminarily set. For example, for a 6205 bearing with a speed of 1800 rpm, the characteristic frequency of the inner race fault is approximately 157 Hz. The cutoff frequency of channel 1 can be initialized to 140-170 Hz, and that of channel 2 to 300-500 Hz to cover harmonics. Channels 3 and 4 are set to a high frequency band (e.g., 1-3 kHz) and a full frequency band (e.g., 10 Hz-5 kHz), respectively, to capture impulse components. Subsequently, a particle swarm optimization (PSO) algorithm was used for joint parameter optimization: the upper and lower cutoff frequencies of each filter (fL1, fH1 to fL4, fH4) were used as particle position vectors. The particle swarm was initialized with 20-50 parameter combinations. The fitness function, F = α·N(Accuracy) + β·N(Confidence), was calculated as the normalized weighted sum of classification accuracy and decision confidence. α and β were adjusted based on the application scenario (default values ​​were α = 0.7 and β = 0.3). Decision confidence was defined as max(y_i) - second_max(y_i), where y_i is the output voltage of the i-th neuron in the simulated neural network. During the optimization process, the PSO algorithm iteratively updates the particle velocity and position. Each iteration configures the current parameters into the hardware circuit (converted to analog voltages by the GP8101 chip to control the MAX261 filter cutoff frequencies). The output of the test set was collected and the fitness value was calculated until the maximum number of iterations (e.g., 100) was reached or the fitness function converged. Finally, the optimal parameter combination that maximizes the F value is saved, including the cutoff frequencies of the four channels (for example, the optimal value for channel 1 is 152-183Hz) and the corresponding neural network weights. This embodiment solves the problem that fixed filters cannot adapt to different bearing models or working conditions through dynamic optimization of hardware parameters. At the same time, by utilizing the real-time processing characteristics of analog circuits, the optimized parameters are directly deployed to edge devices, achieving a balance between diagnostic accuracy and power consumption.

[0070] In an optional embodiment, the working process of the analog neural network classifier includes: inputting the four-dimensional eigenvalues ​​and the corresponding weight parameters into the analog multiplier array according to a predetermined timing through the timing control circuit; inputting the multiplication result into the integration circuit for weighted summation operation; comparing the steady-state voltage values ​​of the four output neurons, and outputting the fault type corresponding to the neuron with the largest voltage value as the diagnosis result.

[0071] In a specific embodiment of the present invention, the hardware implementation process of the analog neural network classifier is as follows: First, the four-dimensional eigenvalues ​​(x1-x4) output by the feature extraction module are synchronized with the pre-optimized weight parameters (wij) by a timing control circuit (such as a 74HC4051 decoder chip) at a timing beat of 0.1ms / channel. The eigenvalues ​​x1-x4 and their corresponding weight parameters w11-w44 of the four neurons are input into the AD633 analog multiplier array in sequence through multiplexing.

[0072] The specific operating sequence is as follows: During the first 0.1ms cycle, x1 is synchronously fed into four multipliers along with w11, w21, w31, and w41. During the second cycle, x2 is fed into four multipliers along with w12-w42, and so on, completing the four sets of multiplication and accumulation. The current signals output by the multipliers are fed into an integrator circuit comprised of an LM741 operational amplifier. The 10nF capacitor and 1MΩ resistor in the feedback loop form a 0.01s time constant, ensuring that the four sets of multiplications are completed within a full 0.5ms timing cycle, achieving the weighted sum ∑xiwij. The integrated outputs of the four neurons are connected to voltage hold circuits (using the LF398 sample-and-hold chip). After the integration process is complete, the steady-state output voltages of the four channels (V1-V4) are compared via an LM339 voltage comparator array. The neuron corresponding to the highest voltage is encoded as a logic "1" (the others are "0"), directly mapping the specific fault type (for example, a "1000" output when V1 is at its maximum indicates a normal state).

[0073] More specifically, the fitness function used in the joint optimization process is:

[0074] F = α·N(Accuracy) + β·N(Confidence)

[0075] Where N(·) represents normalization processing, α and β are adjustable weight coefficients, and α+β=1;

[0076] Decision confidence is defined as:

[0077] Confidence = max(y_i) - second_max(y_i)

[0078] Where y_i represents the output value of the i-th neuron, and second_max(y_i) represents the second largest output value.

[0079] According to the content of the above embodiment, this application proposes a bearing fault diagnosis method based on a multi-channel analog filtering feature network. Its technical principle is to achieve deep integration of signal processing and feature extraction through analog circuit hardware, and combine intelligent optimization algorithms to build an efficient and low-power diagnosis system. The specific workflow is as follows: first, the bearing vibration signal is preprocessed, including high-pass filtering to remove the DC component, amplitude normalization to eliminate dimensional differences, and fixed time window segmentation; the preprocessed signal input is a multi-channel network composed of 4 fourth-order Butterworth bandpass filters, and the passband parameters of each channel are dynamically optimized by the particle swarm algorithm. The weighted sum of classification accuracy and decision confidence is used as the fitness function to ensure the best match between the feature extraction frequency band and the fault characteristics; the root mean square value of the output signal of each channel constitutes a four-dimensional feature vector, which is input into the analog neural network classifier through the timing control circuit, and the multiplication and accumulation operation of the features and the optimized weights is completed through the AD633 multiplier array, and then the weighted sum is realized through the LM741 integration circuit; finally, the steady-state output voltages of the four neurons are compared, and the fault type is output in a coded manner.

[0080] Example 2

[0081] Another specific embodiment of the present application proposes a bearing fault diagnosis system based on a multi-channel analog filtering characteristic network, which is applied to perform the bearing fault diagnosis method in Example 1. The system includes:

[0082] Signal acquisition module, used to obtain original bearing vibration signals;

[0083] Signal preprocessing module, used to denoise and standardize the original signal;

[0084] Multi-channel filtering module, including multiple bandpass filters set in parallel;

[0085] Feature extraction module, used to calculate the characteristic value of the output signal of each channel;

[0086] Simulated neural network classification module for fault classification based on feature values;

[0087] Result display module, used for outputting diagnosis results;

[0088] Among them, the parameters of the multi-channel filtering module and the simulated neural network classification module are pre-configured through the offline optimization module;

[0089] The simulated neural network classification module includes:

[0090] A timing control unit, configured to input the four-dimensional eigenvalues ​​and corresponding weight parameters into the analog multiplier array according to a predetermined timing sequence through a timing control circuit;

[0091] An analog multiplication-addition unit, comprising an analog multiplier array and an integration circuit, inputs the multiplication result into the integration circuit for weighted summation operation;

[0092] The comparison output unit is used to compare the steady-state voltage values ​​of the four output neurons and output the fault type corresponding to the neuron with the largest voltage value as the diagnosis result.

[0093] Furthermore, the feature extraction module uses a fourth-order Butterworth bandpass filter and a root mean square detection circuit to extract four-dimensional features; the result display module is used to compare the output voltages of the four neurons, encode the channel corresponding to the maximum value as 1, and the rest as 0, and output the fault type.

[0094] Furthermore, the system also includes a parameter setting module, which includes:

[0095] Parameter input interface, used to receive filter cutoff frequency and neural network weight parameters;

[0096] A digital-to-analog conversion unit converts digital parameters into corresponding analog voltage signals;

[0097] The parameter distribution unit outputs the converted analog signals to:

[0098] Configure the passband parameters of each bandpass filter in the multi-channel filtering module;

[0099] Simulate the multiplier array of the neural network classification module and provide weight parameters;

[0100] Among them, the working timing of the parameter setting module is synchronized with the timing control unit of the simulated neural network classification module.

[0101] According to the above-mentioned embodiments, the operating principle of the bearing fault diagnosis system based on a multi-channel analog filtering feature network proposed in this application is to achieve real-time, low-power diagnosis of bearing faults through the coordinated cooperation of analog hardware circuits and intelligent optimization algorithms. During system operation, the vibration sensor first collects the raw signal. After the preprocessing module removes the DC component and noise, it is input into the multi-channel filtering module for feature extraction. This module uses four independently adjustable fourth-order Butterworth bandpass filters, whose passband parameters are preconfigured by the offline optimization module to ensure coverage of the characteristic frequency range of bearing faults. The output signal of each filter is converted into an eigenvalue representing the signal energy by the root mean square detection circuit, forming a four-dimensional feature vector. The analog neural network classification module, through the timing control unit, inputs the eigenvalues ​​and optimized weights into the analog multiplier array according to a predetermined time sequence. After the product-accumulation operation is completed, the integrator circuit performs weighted summation. Finally, the comparison output unit detects the steady-state voltage of the four neurons and outputs the fault type code using a "winner-takes-all" mechanism. Throughout this process, the parameter setting module uses digital-to-analog conversion to configure the optimized parameters to each hardware unit in real time, ensuring optimal system performance.

[0102] In another optional embodiment of the present application, an electronic device is proposed, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the bearing fault diagnosis method in Example 1 are implemented.

[0103] In another optional embodiment of the present application, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps of the bearing fault diagnosis method in Example 1 are implemented.

[0104] To further illustrate the bearing fault diagnosis method of the present invention, the above scheme will be further described below in conjunction with some diagnostic cases and processes.

[0105] The present invention uses the bearing signal of Case Western Reserve University for experimental verification, using four types of vibration signals: normal, inner ring 0.178mm fault, ball 0.178mm fault, and outer ring 0.178mm fault; the signal is preprocessed, and the preprocessed data is labeled and divided into training set and test set, which are input into the designed model for classification. The specific design of the model is as follows: the present invention designs four independent fourth-order Butterworth bandpass filters, performs four bandpass filters on each sample data, and calculates the root mean square value of the signal in turn, thereby realizing feature extraction; the four root mean square values ​​obtained are used as the four-dimensional feature input of the linear fully connected layer for classification; the input layer of the linear fully connected layer network is four nodes, which correspond to the four input features in turn, and the output layer is four nodes, which correspond to the four types of faults.

[0106] The optimization algorithm is specifically the particle swarm optimization algorithm, and the implementation process is as follows Figure 2As shown, the filter passband is optimized, and the passband is four different filter cutoff frequencies.

[0107] Optimization index 1 is the accuracy of neural network classification, and optimization index 2 is the decision confidence. The hybrid index of the two is used as the fitness function of the particle swarm optimization algorithm. The decision confidence is The maximum decision result output by a sample signal in four neurons minus the second largest decision result;

[0108]

[0109] in is the output result of the i-th neuron, and;

[0110]

[0111] A hybrid index is used as the fitness function of the optimization algorithm, and the hybrid index is and Normalize them separately and multiply them by weight coefficient to obtain the weighted sum;

[0112]

[0113] in, and They are respectively expressed as the weight ratio of optimization indicators, and They are represented as the normalized results of optimization index 1 and optimization index 2 respectively.

[0114] Decision confidence is used to enhance the robustness of the system. The particle swarm optimization algorithm stops after reaching the maximum number of iterations or optimizing the optimal solution. At this time, four optimal cutoff frequencies of the signal can be obtained. The particle swarm optimization algorithm has a simple principle, fast convergence speed, strong robustness, and easy scalability. It can take into account both global search and local development capabilities and can quickly approach the optimal solution. After the optimization is completed, the model parameters are saved. The model parameters include the cutoff frequencies of the four filters and the weights and bias terms of the linear fully connected layer of the model.

[0115] Design the hardware architecture of the circuit consisting of feature extraction module, simulated neural network classification module and parameter setting module. The overall hardware architecture is implemented as follows Figure 3 As shown; the parameter setting module part is as follows Figure 4 As shown, the PAC chip is used to set the parameters as the corresponding voltage signal. The present invention uses the GP8101 chip to convert the PWM signal into a 0-5V voltage signal, and the voltage signal corresponding to the parameter is input into the digital reasoning unit for decision-making; the feature extraction module filters the signal and calculates the root mean square value. The implementation process is as follows Figure 5As shown, the filter is designed as a fourth-order Butterworth bandpass filter. The present invention uses four MAX261 active filter chips for filtering. The four-channel signals after filtering with four different cutoff frequencies are respectively subjected to RMS calculation to obtain four RMS values ​​as features. The present invention uses four AD637 chips to perform RMS calculation on the signals.

[0116] The simulation neural network classification module is composed of a timing reorganization module, a simulation multiplier and an integration circuit module. The implementation architecture of the simulation neural network classification module is as follows: Figure 6 As shown, the present invention uses five 74HC4051 decoder chips as timing reorganization modules, four 74HC4051 chips are used for the parameters corresponding to each output layer neuron, and one 74CH4051 chip is used to control feature input, controlling the decoder to rearrange the parameters of the trained model and the four-dimensional features of the input signal in a one-to-one correspondence according to the time series; the input features and the weights and biases of each neuron of the model are input into the analog multiplier. The present invention uses an array of four AD633 analog multipliers to perform operations on four neurons respectively. The time series signals obtained after multiplication are input into the integration circuit module for integration operation. The integration circuit module is built using an operational amplifier chip. The operational amplifier in the present invention uses four LM741 chips, and the output corresponding to each neuron is connected to an oscilloscope for observation.

[0117] The different output results of the inference circuit are encoded to achieve fault diagnosis of the bearing signal; the encoding method is that after the integration is completed, the integrator output voltage represents the activation value of the neuron, and the voltage output of the four integration circuits is compared. , , , , identify the highest value among the voltage values ​​output by the integration circuit module, the corresponding neuron output code is 1, and the corresponding output codes of the remaining neurons are 0; the specific encoding method of this case is shown in Table 1:

[0118] Table 1 Correspondence between bearing fault types and output codes

[0119]

[0120] Finally, each type of signal is divided into 58 samples and input into the trained model for inference. The output of the four neurons is observed to determine the type of signal.

[0121] In the hardware circuit framework of this implementation case, the core signal processing chain operates in the analog domain, and only the parameter setting circuit and the timing control circuit involve digital circuits. The GP8101 of the parameter setting circuit is a low-speed DAC, and the 74HC4051 is a low-power digital chip. Compared with embedded devices that rely on high-speed ADCs to collect vibration signals and deploy neural networks for parallel digital calculations, the mixed signal processing architecture of the present invention significantly reduces the overall power consumption of the system and is suitable for scenarios with low power requirements.

[0122] It should be noted that:

[0123] The specific chip models (MAX261, AD637, 74HC4051, AD633, LM741, GP8101) mentioned in the above embodiments are merely preferred examples of implementation solutions and are not intended to limit the present invention. Those skilled in the art may choose other functionally equivalent chips or discrete components to build circuits based on factors such as performance, cost, and availability.

[0124] The above description is merely a preferred embodiment of the present disclosure and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of the present disclosure is not limited to technical solutions formed by specific combinations of the aforementioned technical features. It also encompasses other technical solutions formed by any combination of the aforementioned technical features or their equivalents, without departing from the scope of the above disclosure. For example, a technical solution formed by replacing the aforementioned features with (but not limited to) technical features with similar functions disclosed in this disclosure.

[0125] In addition, although each operation is described in a specific order, this should not be understood as requiring these operations to be performed in the specific order shown or in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details have been included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0126] The above-described embodiments merely illustrate several implementations of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, and all such variations and improvements fall within the scope of protection of the present invention.

Claims

1. A bearing fault diagnosis method based on a multi-channel analog filter characteristic network, characterized in that: The method comprises: Constructing a multi-channel analog filtering feature extraction network, the network comprising a plurality of bandpass filter channels arranged in parallel, each channel being configured with different passband parameters, the passband parameters including a lower cutoff frequency and an upper cutoff frequency; Collect bearing vibration signals and perform signal preprocessing; Inputting the preprocessed signal into the multi-channel analog filtering feature extraction network, and calculating the root mean square value of the output signal of each channel as the feature value; Inputting the characteristic value into a simulated neural network classifier for fault classification; The passband parameters of the bandpass filter channel and the parameters of the simulated neural network classifier are jointly optimized in combination with the classification accuracy and decision confidence; The signal preprocessing includes: The DC component in the original vibration signal is removed by a high-pass filter to obtain an AC vibration signal; performing amplitude normalization processing on the AC vibration signal to obtain a standardized vibration signal; Segmenting the standardized vibration signal according to fixed time windows to obtain signal segments to be analyzed; The multi-channel analog filtering feature extraction network includes four independent fourth-order Butterworth bandpass filter channels, and the upper and lower cutoff frequencies of each channel are determined by the following method: Initialize the cutoff frequency range of each channel based on the characteristic frequency distribution range of the bearing fault; The particle swarm optimization algorithm is used to iteratively optimize the upper and lower cutoff frequencies of each channel, with the weighted sum of classification accuracy and decision confidence as the fitness function; Save the optimal cutoff frequency combination that maximizes the fitness function; The working process of the simulated neural network classifier includes: Inputting the four-dimensional eigenvalues ​​and corresponding weight parameters into the analog multiplier array according to a predetermined timing sequence through a timing control circuit; The multiplication result is input into the integration circuit for weighted summation operation; The steady-state voltage values ​​of the four output neurons are compared, and the fault type corresponding to the neuron with the largest voltage value is output as the diagnosis result.

2. The bearing fault diagnosis method based on multi-channel analog filtering characteristic network according to claim 1 is characterized in that: The fitness function used in the joint optimization process is: F = α·N(Accuracy) + β·N(Confidence) Where N(·) represents normalization processing, α and β are adjustable weight coefficients, and α+β=1; The decision confidence is defined as: Confidence = max(y_i) - second_max(y_i) where y_i represents the output value of the i-th neuron and second_max(y_i) represents the second largest output value.

3. Bearing fault diagnosis system based on multi-channel analog filter characteristic network, characterized by: Applied to executing the bearing fault diagnosis method according to any one of claims 1-2, the system comprises: Signal acquisition module, used to obtain original bearing vibration signals; Signal preprocessing module, used to denoise and standardize the original signal; Multi-channel filtering module, including multiple bandpass filters set in parallel; Feature extraction module, used to calculate the characteristic value of the output signal of each channel; Simulated neural network classification module for fault classification based on feature values; Result display module, used for outputting diagnosis results; The parameters of the multi-channel filtering module and the simulated neural network classification module are pre-configured by an offline optimization module; The simulated neural network classification module includes: A timing control unit, configured to input the four-dimensional eigenvalues ​​and corresponding weight parameters into the analog multiplier array according to a predetermined timing sequence through a timing control circuit; An analog multiplication-addition unit, comprising an analog multiplier array and an integration circuit, inputs the multiplication result into the integration circuit for weighted summation operation; The comparison output unit is used to compare the steady-state voltage values ​​of the four output neurons and output the fault type corresponding to the neuron with the largest voltage value as the diagnosis result.

4. The bearing fault diagnosis system based on multi-channel analog filtering characteristic network according to claim 3 is characterized in that: The feature extraction module uses a fourth-order Butterworth bandpass filter and a root mean square detection circuit to extract four-dimensional features; the result display module is used to compare the output voltages of four neurons, encode the channel corresponding to the maximum value as 1, the others as 0, and output the fault type.

5. The bearing fault diagnosis system based on multi-channel analog filtering characteristic network according to claim 4 is characterized in that: The system further includes a parameter setting module, which includes: Parameter input interface, used to receive filter cutoff frequency and neural network weight parameters; A digital-to-analog conversion unit converts digital parameters into corresponding analog voltage signals; The parameter distribution unit outputs the converted analog signals to: Each bandpass filter of the multi-channel filtering module is configured with its passband parameters; The multiplier array of the simulated neural network classification module provides weight parameters; Among them, the working timing of the parameter setting module is synchronized with the timing control unit of the simulated neural network classification module.

6. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the bearing fault diagnosis method according to any one of claims 1 to 2 when executing the computer program.

7. A computer-readable storage medium, characterized in that A computer program is stored, and when the computer program is executed by a processor, the steps of the bearing fault diagnosis method according to any one of claims 1 to 2 are implemented.

Images