Gated impulse neuron and dual-channel simulation based bearing fault diagnosis system

Abstract

The application belongs to the technical field of bearing fault diagnosis, and particularly relates to a bearing fault diagnosis system based on a gating pulse neuron and a double-channel simulation, which comprises an offline parameter optimization unit arranged on an upper computer and an online simulation reasoning unit arranged on an edge end. The offline parameter optimization unit generates filter parameters and decision thresholds of a first channel according to inter-class distance between outer ring fault and other states of bearing vibration data, and generates filter parameters and decision thresholds of a second channel according to inter-class distance between normal state and other states. Each channel of the online simulation reasoning unit sequentially filters, extracts features and compares voltages of signals, and turns on a power supply to drive neurons to emit pulses when the feature voltage exceeds the threshold value. The system outputs a diagnosis result according to a double-channel pulse combination. The application solves the problem of mismatch between weak feature voltage and neuron excitation threshold, and realizes low-power bearing fault diagnosis.

Description

Technical Field

[0001] This invention belongs to the field of bearing fault diagnosis technology, specifically relating to a bearing fault diagnosis system based on gated pulse neurons and dual-channel simulation. Background Technology

[0002] Currently, the mainstream technical solution for rolling bearing fault diagnosis adopts a high-frequency analog-to-digital converter (ADC) plus digital signal processing (DSP) architecture. Specifically, an ADC converts the analog signals collected by vibration sensors into digital signals, and a DSP then uses algorithms such as Fast Fourier Transform (FFT), wavelet packet decomposition, and support vector machines (SVM) to extract features and classify faults. This solution has been widely used in industrial equipment condition monitoring. With the development of the Industrial Internet of Things (IIoT), some research attempts to port lightweight diagnostic models to embedded devices to achieve real-time monitoring at the device level.

[0003] However, the above-mentioned technical solutions have the following problems in practical applications. First, high-frequency analog-to-digital conversion and digital signal processing require continuous large-scale data computation, resulting in high system power consumption and failing to meet the requirements of long-term maintenance-free deployment. Second, digital processing solutions suffer from computational latency and data transfer overhead, which is insufficient for fault early warning scenarios requiring rapid response. Third, existing lightweight models ported to the edge need to make a trade-off between diagnostic accuracy and power consumption, and still rely on analog-to-digital conversion and digital computation. Fourth, although pure analog diagnostic circuits have low power consumption, their fixed filter parameters cannot adapt to changes in fault frequency bands under different operating conditions, and the diagnostic effect decreases when the equipment speed or load changes. Fifth, the DC voltage amplitude obtained after weak fault signals are collected by sensors is very small, making it difficult to directly drive the subsequent pulse excitation circuit, resulting in fault characteristics not being effectively identified or causing erroneous triggering. Summary of the Invention

[0004] The purpose of this invention is to provide a bearing fault diagnosis system based on gated spiking neurons and dual-channel simulation, in order to solve the problems in existing bearing fault diagnosis where weak characteristic voltages cannot effectively drive simulated spiking neurons, and where diagnostic circuit parameters cannot adapt to changes in operating conditions.

[0005] The present invention achieves the above objectives through the following technical solutions: This invention proposes a bearing fault diagnosis system based on gated spiking neurons and dual-channel simulation, comprising an offline parameter optimization unit deployed on the host computer and a simulation inference unit deployed at the edge. The online simulation inference unit includes a first channel and a second channel; the first channel is used to extract the vibration energy characteristics of the sensitive frequency band of the bearing outer ring fault; the second channel is used to extract the vibration energy characteristics of the sensitive frequency band of the bearing inner ring fault, so as to distinguish it from the normal state; in addition, since the outer ring fault has wide frequency domain strong impact characteristics, it exhibits high energy characteristics in the sensitive frequency band of the dual channels.

[0006] The offline parameter optimization unit is used to generate the first optimal filter parameters and the first decision threshold of the first channel based on the pre-collected multi-state vibration data of the bearing, with the goal of maximizing the distance between the outer ring fault and other state classes, and to generate the second optimal filter parameters and the second decision threshold of the second channel with the goal of maximizing the distance between the normal state and other state classes. The first and second channels sequentially perform filtering, feature extraction, voltage comparison gating and pulse generation. When the feature voltage exceeds the corresponding threshold, the independent power supply is turned on to trigger the discharge pulse, forming a pulse emission state combination. The corresponding bearing operation classification result is determined in response to the pulse delivery state combination.

[0007] Furthermore, the offline parameter optimization unit uses the particle swarm optimization algorithm to jointly optimize the filter parameters of the first channel and the second channel, and constructs a fitness function with a smooth guided penalty mechanism. Inter-class spacing expectation for the first channel The calculation formula is: Inter-class spacing expectation for the second channel The calculation formula is: in, , , These represent the root mean square feature centers of the outer ring fault, normal state, and inner ring fault extracted after filtering through the first channel, respectively. , , These represent the root mean square feature centers of the inner circle fault, outer circle fault, and normal state extracted after filtering via the second channel, respectively. A linear penalty is applied when category overlap occurs in the feature space; when the overlap is eliminated, i.e., ... and When the fitness is calculated, the formula is: .

[0008] Furthermore, the online simulation inference unit also includes a microcontroller, and both the first and second channels include a programmable switched-capacitor filter, a true RMS conversion chip, a linear operational amplifier circuit, a voltage comparator, an analog switch, and an analog spiking neuron circuit; wherein: The programmable switched-capacitor filter performs filtering on the input signal; The true RMS conversion chip and linear operational amplifier circuit perform feature extraction on the filtered signal to generate an amplified DC characteristic voltage. The voltage comparator and analog switch perform voltage comparison gating. When the DC characteristic voltage exceeds the corresponding decision threshold, the analog switch turns on the independent constant drive power supply. The simulated spiking neuron circuit responds to the power supply being turned on to generate a pulse to trigger a discharge pulse, forming a pulse firing state combination of the first channel and the second channel.

[0009] Furthermore, the offline parameter optimization unit and the online simulation inference unit adopt a communication mode of one-time configuration or periodic update; when there is no update instruction, the online simulation inference unit independently performs pure hardware diagnosis based on the configured parameters, without needing to maintain real-time communication with the host computer.

[0010] Furthermore, the method of determining the corresponding bearing operation classification result in response to the pulse firing state combination includes: When neither the first nor the second channel issues a pulse, the response pulse issuance state combination (0,0) determines the corresponding bearing operation classification result as normal operation state; When only the second channel issues a pulse and the first channel does not issue a pulse, the response pulse issuance state combination (0,1) determines the corresponding bearing operation classification result as inner ring fault; When both the first and second channels emit pulses, the response pulse emission state combination (1,1) determines the corresponding bearing operation classification result as outer ring fault.

[0011] Furthermore, the simulated spiking neuron circuit is constructed using discrete transistors and an RC network, including an input coupling resistor connected in series on the input path, a film capacitor and a leakage resistor connected in parallel to ground, a positive feedback depolarization loop composed of complementary NPN and PNP transistors, and a repolarization refractory period loop composed of a delay capacitor; the voltage value of the constant drive power supply of the system is higher than the inherent excitation threshold of the positive feedback depolarization loop.

[0012] Furthermore, the signal input terminal of the voltage comparator receives the amplified DC characteristic voltage, and the reference terminal is connected to the corresponding decision threshold that is amplified proportionally; the control terminal of the analog switch is connected to the output terminal of the voltage comparator, the signal input terminal is independently connected to the system's constant drive power supply, and the signal output terminal is connected to the analog spiking neuron circuit.

[0013] This invention also proposes a bearing fault diagnosis method based on gated spiking neurons and dual-channel simulation, comprising the following steps: S1: Collect the real-time vibration acceleration signal of the bearing, perform mean zeroing preprocessing, and then divide it into segments according to the preset length to construct a sample dataset; S2: In the host computer, the particle swarm optimization algorithm is used to jointly optimize the filter parameters of the first channel and the second channel. The optimal center frequency, bandwidth and first decision threshold of the first channel are output with the goal of maximizing the distance between the outer ring fault and other state classes. The optimal center frequency, bandwidth and second decision threshold of the second channel are output with the goal of maximizing the distance between the normal state and other state classes. S3: The optimal center frequency and bandwidth are sent to the microcontroller at the edge. The microcontroller dynamically configures the programmable switched capacitor filter chip to physically filter the input vibration signal and remove background noise. S4: Input the filtered AC signal into the true RMS converter chip to convert it into a DC characteristic voltage, and then amplify it through the linear operational amplifier circuit to output the amplified DC characteristic voltage. S5: The amplified DC characteristic voltage is input to the voltage comparator and compared with the corresponding decision threshold after proportional amplification; if the DC characteristic voltage is greater than the decision threshold, the voltage comparator outputs a high level to drive the analog switch to turn on, and the independent constant drive power supply is released to the analog spiking neuron circuit; if the DC characteristic voltage is not greater than the decision threshold, the analog switch remains off; S6: When the simulated spiking neuron circuit receives a conducting driving power supply, the internal membrane capacitor depolarizes and emits a pulse sequence; when it does not receive a driving power supply, it remains silent. S7: Detects the pulse output status of the first and second channels. When there is no pulse in both channels, the output (0,0) indicates normal operation. When there is a pulse only in the second channel, the output (0,1) indicates an inner ring fault. When there is a pulse in both channels, the output (1,1) indicates an outer ring fault.

[0014] Furthermore, the decision threshold mentioned in step S5 is amplified and mapped proportionally on the hardware side according to the same amplification factor as the DC characteristic voltage.

[0015] The beneficial effects of this invention are as follows: 1. This invention generates optimal filter parameters and decision thresholds offline using a particle swarm optimization algorithm on the host computer, and then sends them down to the edge to configure a programmable filter chip. This allows the same hardware to adapt to different bearing models and fault frequency band changes under different operating conditions, solving the problems of fixed parameters and poor generalization ability of pure analog diagnostic circuits.

[0016] 2. In this invention, the entire edge diagnostic process uses pure analog circuits to complete filtering, feature extraction, and gating comparison, eliminating the need for analog-to-digital converters and digital signal processors, thus significantly reducing system power consumption. Furthermore, a gating circuit composed of a voltage comparator and analog switches isolates weak DC characteristic voltages from the subsequent spiking neurons. Only when the characteristic voltage exceeds a decision threshold is the independent driving power supply activated to trigger the neuron to fire pulses, solving the problems of weak characteristic voltages not being able to directly drive spiking neurons and the susceptibility to false triggering near the critical point.

[0017] 3. This invention directly outputs diagnostic results based on the pulse firing combination of dual-channel neurons, without the need for complex digital calculations, resulting in fast response speed and strong anti-interference ability. Attached Figure Description

[0018] Figure 1 This is a system block diagram of the bearing fault diagnosis system in an embodiment of the present invention; Figure 2 This is a flowchart of a bearing fault diagnosis method in an embodiment of the present invention; Figure 3 This is another flowchart of the bearing fault diagnosis method in an embodiment of the present invention; Figure 4 This is a schematic diagram of the threshold gate control unit circuit in an embodiment of the present invention; Figure 5 This is a schematic diagram of the simulated spiking neuron network circuit in an embodiment of the present invention; Figure 6 This is an oscilloscope waveform of the output pulse measured by the hardware system when the input is in a normal state, as used in the experimental verification of this invention. Figure 7 This is an oscilloscope waveform of the hardware system measured when the inner ring fault characteristics are input during the experimental verification of this invention. Figure 8 This is an oscilloscope waveform of the output pulse of the hardware system when the outer ring fault characteristics are input during the experimental verification of this invention. Detailed Implementation

[0019] The following description provides specific application scenarios and requirements for this specification, intended to enable those skilled in the art to make and use the contents of this specification. Various partial modifications to the disclosed embodiments will be apparent to those skilled in the art, and the general principles defined herein can 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 rather to the widest scope consistent with the claims.

[0020] The terminology used herein is for the purpose of describing particular exemplary embodiments only and is not restrictive. For example, unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “the” used herein may also include the plural forms. When used in this specification, the terms “comprising,” “including,” and / or “containing” mean that the associated integers, steps, operations, elements, and / or components are present, but do not exclude the presence of one or more other features, integers, steps, operations, elements, components, and / or groups, or that other features, integers, steps, operations, elements, components, and / or groups may be added to the system / method.

[0021] Considering the following description, these and other features of this specification, as well as the operation and function of the related components of the structure, and the economy of assembly and manufacture of the parts, can be significantly improved. All of these form part of this specification with reference to the accompanying drawings. However, it should be clearly understood that the drawings are for illustrative and descriptive 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.

[0022] The flowcharts used in this specification illustrate operations implemented according to some embodiments of this specification. It should be clearly understood that the operations in the flowcharts may not be implemented in a sequential order. Instead, 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.

[0023] First Embodiment Combination Figure 1 This embodiment provides a bearing fault diagnosis system based on gated spiking neurons and dual-channel simulation. The system is divided into two parts: an offline parameter optimization unit deployed on a host computer, and an online simulation inference unit deployed at the edge. The host computer and the edge unit interact via a communication connection. In practical deployment, the host computer can be a personal computer, industrial control computer, or cloud server, used to perform computationally intensive parameter optimization tasks; the edge unit is deployed at the equipment site, close to the vibration sensors, to perform lightweight real-time diagnostic tasks.

[0024] The online simulation inference unit includes a first channel and a second channel. The first channel is used to extract the vibration energy characteristics of the sensitive frequency band for outer ring faults, and the second channel is used to extract the vibration energy characteristics of the sensitive frequency band for distinguishing between inner ring faults and normal conditions. The offline parameter optimization unit is used to generate the first optimal filter parameters and the first decision threshold for the first channel based on the pre-collected multi-state vibration data of the bearing, with the goal of maximizing the distance between outer ring faults and other state classes, and to generate the second optimal filter parameters and the second decision threshold for the second channel with the goal of maximizing the distance between normal conditions and other state classes. The first and second channels sequentially perform filtering, feature extraction, voltage comparison gating, and pulse generation. When the characteristic voltage exceeds the corresponding threshold, an independent power supply is turned on to trigger a discharge pulse, forming a pulse emission state combination. The corresponding bearing operation classification result is determined in response to the pulse emission state combination.

[0025] Specifically, the bearing fault diagnosis system proposed in this invention will be analyzed in detail below with reference to the accompanying drawings.

[0026] Combination Figure 1 and Figure 2 The online simulation inference unit includes a first channel and a second channel, which operate in parallel. The first channel is used to extract the vibration energy characteristics of the sensitive frequency band for outer ring faults in the bearing, while the second channel is used to extract the vibration energy characteristics of the sensitive frequency band for distinguishing between inner ring faults and normal conditions. The filter parameters and decision thresholds for the two channels are determined independently by the offline parameter optimization unit.

[0027] (I) Specific implementation of the offline parameter optimization unit The offline parameter optimization unit employs a particle swarm optimization (PSO) algorithm to jointly optimize the filter parameters of the first and second channels. During the optimization process, the system first needs to pre-collect vibration data of the bearing under different operating conditions, including normal conditions, inner race fault conditions, and outer race fault conditions. During data acquisition, the sampling frequency can be set to 12kHz, and a sufficient number of data samples are collected for each condition, for example, more than 100 sets of samples for each condition, with each set containing 2048 data points.

[0028] To overcome the shortcomings of traditional methods that overly rely on prior bearing geometric parameters and are highly susceptible to frequency band drift caused by complex and varying operating conditions, this invention employs a purely data-driven approach to adaptively find the optimal fault frequency band. The specific execution logic is as follows: Two parallel bandpass filters are designed as the first channel (defined as the X-axis feature distinguishing outer ring faults) and the second channel (defined as the Y-axis feature distinguishing normal from faulty). A particle swarm optimization algorithm is used to globally and jointly optimize the combination of the center frequency and bandwidth parameters of these two filters, and the bandwidth is forced to be no less than 1000Hz to prevent narrowband overfitting.

[0029] The fitness function is constructed by introducing a smoothing-guided penalty mechanism. For the filtering parameters corresponding to each particle, the root mean square (RMS) feature distributions of the three types of states in the training set after filtering are calculated respectively.

[0030] Inter-class spacing expectation for the first channel The calculation formula is: Inter-class spacing expectation for the second channel The calculation formula is: in, , , These represent the root mean square feature centers of the outer ring fault, normal state, and inner ring fault extracted after filtering through the first channel, respectively. This represents a function that takes the maximum value in the set. The first channel aims to separate outer ring faults from other states, therefore it is desirable that the characteristic mean of outer ring faults be as large as possible compared to the other two classes. , , These represent the root mean square feature centers of the inner circle fault, outer circle fault, and normal state extracted after filtering via the second channel, respectively. The function represents the minimum value in the set. The second channel aims to separate the normal state from the other states, therefore it is desirable that the characteristic mean of the normal state is as small as possible compared to the other two classes.

[0031] When class overlap occurs in the feature space, a linear penalty value is applied to guide particles to gradually escape the blind zone. When class separation is successful, and the overlap is eliminated... and When the fitness is calculated, the formula is: .

[0032] Where Fitness is the fitness function evaluation value in the particle swarm optimization algorithm.

[0033] The system iteratively optimizes and ultimately determines the optimal dual-channel filter bandwidth parameters, namely the optimal center frequency of the first channel. With bandwidth and the optimal center frequency of the second channel. With bandwidth Subsequently, based on the non-intersecting two-dimensional feature space after classification, the midpoint of the geometric center of each cluster is extracted, and the first channel determination threshold is adaptively generated. With the second channel judgment threshold .

[0034] (II) Specific Implementation of Online Simulated Reasoning Unit The online analog inference unit includes a microcontroller (such as an STM32 series chip) and hardware modules shared by the first and second channels or configured independently. Each channel includes a programmable switched-capacitor filter (such as a MAX261 chip), a true RMS converter chip (such as an AD637 chip), a linear operational amplifier circuit, a voltage comparator (such as an LM393 chip), an analog switch (such as a CD4066 chip), and an analog spiking neuron circuit.

[0035] Before system operation, the microcontroller receives the optimal filter parameters (including center frequency and bandwidth) from the offline parameter optimization unit and dynamically configures the programmable switched-capacitor filter chip. For example, the microcontroller generates a specific frequency-divided clock signal and writes it into the MAX261 chip, making it form a physical bandpass filter that strictly corresponds to the target frequency band. This configuration can be a one-time event, i.e., configured during system deployment; or it can be periodic, i.e., the configuration is updated after the host computer re-optimizes based on new operating data. When there is no update instruction, the online simulation inference unit independently performs pure hardware diagnostics based on the configured parameters, without needing to maintain real-time communication with the host computer, which reduces the system's dependence on the communication network.

[0036] like Figure 2 As shown, the workflow of the online simulation inference unit is as follows: First, a programmable switched-capacitor filter filters the input bearing vibration signal, physically removing background noise across the entire frequency band and separating the analog AC signal from the specific fault-sensitive frequency band. This filter is dynamically configured by a microcontroller, capable of precisely setting the center frequency and bandwidth based on offline optimization results, thus adapting to changes in the fault frequency band under different bearing models and operating conditions.

[0037] Secondly, the true RMS converter chip receives the filtered analog AC signal, performs squaring and integration operations on it in the pure analog domain, and outputs a DC characteristic voltage representing the vibration energy of that frequency band. The true RMS converter chip internally contains a squarer, integrator, and square root extraction circuit, capable of converting the RMS value of the AC signal into a DC level. However, the DC characteristic voltage output by this chip is typically in the sub-volt range (e.g., 0.05V to 0.8V), with a small amplitude, making it susceptible to interference from circuit board ground noise.

[0038] Therefore, in the linear operational amplifier circuit, the original DC characteristic voltage is precisely amplified, for example, by a factor of 10, to obtain an amplified DC characteristic voltage with a high signal-to-noise ratio. Simultaneously, the first decision threshold and the second decision threshold obtained in the offline algorithm stage are also amplified by the same factor (for example, by a factor of 10) and used as the actual reference threshold voltage for the comparator in the physical hardware circuit.

[0039] Then, the voltage comparator receives the amplified DC characteristic voltage, and its reference terminal is connected to the corresponding decision threshold, which is amplified proportionally. The control terminal of the analog switch is connected to the output terminal of the voltage comparator, the signal input terminal is independently connected to the system's constant drive power supply (e.g., +5V DC power supply), and the signal output terminal is connected to the analog spiking neuron circuit. The design of this gated circuit cleverly isolates the low-voltage analog computation domain from the high-voltage neuron excitation domain. When the amplified DC characteristic voltage is greater than the decision threshold, the voltage comparator outputs a high level, driving the analog switch to conduct, allowing the independent constant drive power supply to the analog spiking neuron circuit; when the amplified DC characteristic voltage is not greater than the decision threshold, the analog switch remains off, and the neuron circuit has no power input.

[0040] Finally, the analog spiking neuron circuit is constructed using discrete transistors and an RC network. For example... Figure 5 As shown, the circuit specifically includes: an input coupling resistor connected in series on the input path, a film capacitor and a leakage resistor connected in parallel to ground, a positive feedback depolarization loop composed of complementary NPN and PNP transistors, and a repolarization refractory period loop composed of a delay capacitor. The voltage value of the system's constant drive power supply (e.g., +5V) is higher than the inherent excitation threshold of the positive feedback depolarization loop (e.g., approximately 0.7V).

[0041] When the analog switch is turned on, a constant +5V drive power supply is applied to the input of the neuron circuit. Because this drive power supply voltage is much higher than the transistor's turn-on threshold and has strong driving capability, it can overcome the effects of input coupling resistance and leakage resistance, rapidly charging the membrane capacitor (i.e., depolarizing it). When the membrane capacitor voltage reaches the trigger threshold of the positive feedback loop, the NPN and PNP transistors quickly enter saturation conduction, generating a large-amplitude pulse at the output. Subsequently, the delay capacitor begins charging, putting the circuit into a refractory period during which the input signal is ignored. After the refractory period ends, the circuit returns to the ready-to-trigger state and can respond to the next drive power supply turn-on event. This circuit has the advantage of extremely low static power consumption because when there is no drive power input, all transistors are in the off state, almost no current flows, and the circuit remains in a silent linear state.

[0042] Combined with appendix Figure 4 and attached Figure 5 The isolated threshold gating unit and the analog spiking neuron circuit in this invention will be further described in detail below: First, such as Figure 4 As shown, the isolated threshold gating unit is composed of a voltage comparator (such as LM393) and an analog switch (such as 4066BD). Specifically, the non-inverting input of the voltage comparator receives the amplified DC characteristic voltage. The inverting input is connected to the corresponding decision threshold, which is amplified proportionally. Since conventional voltage comparators are mostly open-drain output structures, this invention includes a pull-up resistor R2 at its output terminal, connected to the system power supply (VDD), to ensure an absolutely steep and stable output control level. The control terminal of the analog switch is connected to the output terminal of the voltage comparator. Its signal input terminal is not connected in series with the characteristic signal of the preceding stage, but is independently connected to the system's constant drive power supply (e.g., +5V DC power supply). The signal output terminal... It is connected to an analog spiking neuron circuit. Furthermore, at the signal output... A pull-down bleed resistor R1 is preferably configured between the gate and ground (GND). The design of this gate circuit cleverly isolates the low-voltage analog computing domain from the high-voltage neuron excitation domain in terms of physical circuitry.

[0043] like Figure 5 As shown, the analog spiking neuron circuit adopts an analog oscillator topology built from discrete components, specifically including: an input coupling resistor connected in series in the input path. Film capacitor connected in parallel to ground With leakage resistance A positive feedback depolarization circuit composed of NPN and PNP transistors, and a delay capacitor. This forms a complex polarization refractory period loop. The input terminal of this circuit is connected to... Figure 4 Output Signal.

[0044] (III) Output of classification results The system outputs the bearing operation classification result based on the pulse firing states of the first and second channels. The logic decoding module monitors the output ports of the two channels' analog spiking neuron circuits in real time: a high-level pulse is recorded as logic "1", and a low-level silent state is recorded as logic "0".

[0045] The specific judgment rules are as follows: When neither the first nor the second channel outputs a pulse, meaning both channels output a low level (corresponding to logic state (0,0), the system is considered to be in normal operation. This is because, under normal conditions, the vibration energy in both frequency bands is low, and the characteristic voltages do not exceed their respective decision thresholds.

[0046] When only the second channel emits a pulse while the first channel does not, corresponding to logic state (0,1), the system determines it to be an inner-circle fault. This is because an inner-circle fault generates stronger vibrational energy in the sensitive frequency band corresponding to the second channel, while the energy is weaker in the frequency band corresponding to the first channel.

[0047] When both the first and second channels emit pulses, corresponding to logic state (1,1), the system determines it to be an outer ring fault. This is because an outer ring fault generates strong vibrational energy in the sensitive frequency bands of both channels, causing the characteristic voltages of both channels to exceed the decision threshold.

[0048] It should be noted that the logic state (1,0) will not appear in the three-classification scheme of this invention, and there is no fault type where only the first channel exceeds the limit while the second channel does not. This asymmetric decision logic design simplifies the classifier structure and improves diagnostic reliability.

[0049] Second Embodiment This embodiment provides a bearing fault diagnosis method based on gated spiking neurons and dual-channel simulation. This method can be implemented using the system described in the first embodiment. Figure 2 and Figure 3 As shown, the method includes the following steps: S1: Acquire real-time vibration acceleration signals from the bearing, perform mean-zeroing preprocessing, and then segment the data into pre-defined lengths to construct a sample dataset. Specifically, a piezoelectric accelerometer can be used to acquire the vibration signals, with a sampling frequency set to 12kHz. Each sample group contains 2048 data points. Mean-zeroing preprocessing is used to eliminate DC bias and avoid affecting subsequent RMS feature extraction.

[0050] S2: In the host computer, a particle swarm optimization algorithm is used to jointly optimize the filter parameters of the first and second channels. The optimal center frequency, bandwidth, and first decision threshold of the first channel are output with the objective of maximizing the distance between the outer ring fault and other state classes; the optimal center frequency, bandwidth, and second decision threshold of the second channel are output with the objective of maximizing the distance between the normal state and other state classes. The specific optimization process and formulas are described in the first embodiment.

[0051] S3: The optimal center frequency and bandwidth are sent to the microcontroller at the edge. The microcontroller dynamically configures the programmable switched capacitor filter chip to perform physical filtering on the input vibration signal and remove background noise.

[0052] S4: The filtered AC signal is input to the true RMS converter chip, converted into a DC characteristic voltage, and then amplified by a linear operational amplifier circuit before being output as the amplified DC characteristic voltage. For example, the amplification factor can be set to 10 times to amplify the original 0.05V-0.8V DC characteristic voltage to 0.5V-8V, which is convenient for subsequent comparator identification.

[0053] S5: Input the amplified DC characteristic voltage into the voltage comparator and compare it with the corresponding decision threshold that has been amplified proportionally.

[0054] It's important to note that the decision threshold needs to be scaled up proportionally in the hardware to the same amplification factor as the DC characteristic voltage. For example, if the first decision threshold generated by the offline algorithm is 0.3V, and the amplification factor is 10, then the reference threshold of the hardware comparator should be set to 3.0V. If the DC characteristic voltage is greater than the decision threshold, the voltage comparator outputs a high level to drive the analog switch to turn on, allowing an independent constant drive power supply (e.g., +5V) to the analog spiking neuron circuit; if the DC characteristic voltage is not greater than the decision threshold, the analog switch remains off.

[0055] S6: When the simulated spiking neuron circuit receives a driving power supply, its internal membrane capacitor depolarizes and emits a pulse sequence; when no driving power supply is received, it remains silent. The hardware implementation of this step is described in the simulated spiking neuron circuit description in the first embodiment.

[0056] S7: Detects the pulse output status of the first and second channels. When there is no pulse in both channels, the output (0,0) indicates normal operation. When there is a pulse only in the second channel, the output (0,1) indicates an inner ring fault. When there is a pulse in both channels, the output (1,1) indicates an outer ring fault.

[0057] Through the above steps, a complete diagnostic process from vibration signal acquisition to fault category output is realized. The entire process does not require analog-to-digital conversion and digital calculation at the edge, with extremely low power consumption and fast response speed.

[0058] Experimental verification To fully demonstrate the industrial applicability and extremely high classification accuracy of the present invention, this embodiment uses the 12kHz sampling rate measured bearing dataset published by Case Western Reserve University for hardware-in-the-loop verification (including three typical operating states: normal, inner race, and outer race).

[0059] In the algorithm optimization and parameter configuration stage: After optimization using the PSO algorithm on the host computer, the optimal filtering parameters for both channels are obtained and sent to the MAX261. For the 10x linear amplification factor of the operational amplifier in the hardware circuit, the system synchronously and proportionally amplifies the actual hardware reference threshold parameter of the gate comparator by 10x for mapping calibration: the actual hardware threshold calibration for the first channel is as follows: (Corresponding to the original threshold of the algorithm: 0.3V); the actual hardware threshold calibration for the second channel is... (Corresponding to the original threshold of 0.4V in the algorithm). The isolation drive voltage for the analog switch is set to a stable and sufficient +5V DC.

[0060] In the classification logic mapping stage: Relying on the above-mentioned optimal frequency band and proportional mapping dual threshold gating mechanism, the system executes the hardware physical mapping logic shown in Table 1: Table 1 System Hardware Physical Diagnosis Mapping Logic Table During the hardware physical testing phase: A dual-channel hardware circuit prototype composed of corresponding discrete electronic components was constructed. The three types of datasets mentioned above were input in real time, and a dual-channel digital oscilloscope was used to capture the voltage changes of the output layer neurons. The real-time dynamic test results are as follows: Figures 6-8 As shown: (1) As Figure 6 As shown, when the input is in normal state, the AD637 extracts only about 0.05V of the dual-channel raw RMS voltage. After being amplified 10 times... The voltage is approximately 0.5V, far below the dual-channel hardware comparison thresholds of 4.0V and 3.0V. With the dual comparator outputs low, the analog switches are firmly shut off. The neurons are physically isolated from the preceding stage; without power input, the dual channels exhibit a near-zero voltage low-level silent line, demonstrating excellent system noise immunity and robust mapping logic "00".

[0061] (2) Figure 7 As shown, when the inner ring fault characteristic is input, the measured DC characteristic voltage after amplification of the first channel is approximately 3.2V, which is blocked by the 4.0V threshold, and the first channel neuron remains silent; while the DC characteristic voltage after amplification of the second channel is approximately 3.4V, successfully exceeding the 3.0V hardware threshold. The comparator flips, driving the analog switch of channel two to conduct, instantly injecting a +5V isolated high-voltage level. The second channel neuron quickly depolarizes and generates continuous periodic high-voltage spike pulses, and the system accurately maps logic "01".

[0062] (3) Figure 8 As shown, when the outer ring fault characteristics are input, the severe damage to the outer ring generates strong energy broadband impulses in both frequency bands. The measured DC characteristic voltage after dual-channel amplification jumps to approximately 8.7V, significantly and clearly exceeding the dual-side decision threshold. With both analog switches simultaneously turned on, both neurons receive a sufficient +5V constant drive from the power network, causing both channels to oscillate simultaneously and generate high-amplitude spike pulses, achieving full-node ignition and perfectly mapping logic "11".

[0063] The above experimental results fully demonstrate that this invention not only overcomes the bias blind spot of theoretical prior calculation in the front-end feature extraction stage, but more importantly, it achieves pure physical dimensionality reduction at the edge hardware level by utilizing programmable chips, hardware-level RMS extraction, and operational amplifier networks, and innovatively introduces a threshold-gated relay valve connected to an independent driving power supply. This mechanism eliminates the level mismatch and insufficient load gap between weak analog features and the neuron oscillation threshold, endowing the originally easily interfered analog hardware with digital-like absolute determinism. Experimental results prove that the system guarantees a bearing fault classification accuracy of over 99% under various operating conditions, completely opening up a crucial link in the transition of ultra-low-power neuromorphic intelligent hardware from theoretical models to harsh industrial physical field implementation, and possesses extremely broad application prospects.

[0064] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0065] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0066] The embodiments described above are merely examples of several implementations of the present invention, and 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 those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A bearing fault diagnosis system based on gated spiking neurons and dual-channel simulation, characterized in that, An offline parameter optimization unit deployed on the host computer, and an online simulation inference unit deployed at the edge; The online simulation inference unit includes a first channel and a second channel; the first channel is used to extract the vibration energy characteristics of the sensitive frequency band of the bearing outer ring fault, and the second channel is used to extract the vibration energy characteristics of the sensitive frequency band that distinguishes between the inner ring fault and the normal state. The offline parameter optimization unit is used to generate the first optimal filter parameters and the first decision threshold of the first channel based on the pre-collected multi-state vibration data of the bearing, with the goal of maximizing the distance between the outer ring fault and other state classes, and to generate the second optimal filter parameters and the second decision threshold of the second channel with the goal of maximizing the distance between the normal state and other state classes. The first and second channels sequentially perform filtering, feature extraction, voltage comparison gating and pulse generation. When the feature voltage exceeds the corresponding threshold, the independent power supply is turned on to trigger the discharge pulse, forming a pulse emission state combination. The corresponding bearing operation classification result is determined in response to the pulse firing state combination, wherein: When neither the first nor the second channel issues a pulse, the response pulse issuance state combination (0,0) determines the corresponding bearing operation classification result as normal operation state; When only the second channel issues a pulse and the first channel does not issue a pulse, the response pulse issuance state combination (0,1) determines the corresponding bearing operation classification result as inner ring fault; When both the first and second channels issue pulses, the response pulse issuance state combination (1,1) determines the corresponding bearing operation classification result as outer ring fault; The online simulation inference unit further includes a microcontroller, and both the first and second channels include a programmable switched-capacitor filter, a true RMS conversion chip, a linear operational amplifier circuit, a voltage comparator, an analog switch, and an analog spiking neuron circuit; wherein: The programmable switched-capacitor filter performs filtering on the input signal; The true RMS conversion chip and linear operational amplifier circuit perform feature extraction on the filtered signal to generate an amplified DC characteristic voltage. The voltage comparator and analog switch perform voltage comparison gating. When the DC characteristic voltage exceeds the corresponding decision threshold, the analog switch turns on the independent constant drive power supply. The simulated spiking neuron circuit responds to the power supply being turned on to generate a pulse to trigger a discharge pulse, forming a pulse firing state combination of the first channel and the second channel; The analog spiking neuron circuit is constructed using discrete transistors and an RC network, including an input coupling resistor connected in series on the input path, a film capacitor and a leakage resistor connected in parallel to ground, a positive feedback depolarization loop composed of complementary NPN and PNP transistors, and a repolarization refractory period loop composed of a delay capacitor; the voltage value of the constant drive power supply of the system is higher than the inherent excitation threshold of the positive feedback depolarization loop.

2. The bearing fault diagnosis system according to claim 1, characterized in that, The offline parameter optimization unit uses the particle swarm optimization algorithm to jointly optimize the filter parameters of the first and second channels, and constructs a fitness function with a smooth guided penalty mechanism. Inter-class spacing expectation for the first channel The calculation formula is: Inter-class spacing expectation for the second channel The calculation formula is: in, , , These represent the root mean square feature centers of the outer ring fault, normal state, and inner ring fault extracted after filtering through the first channel, respectively. , , These represent the root mean square feature centers of the inner circle fault, outer circle fault, and normal state extracted after filtering via the second channel, respectively. When category overlap occurs in the feature space, i.e. or When a linear penalty value is applied, the fitness calculation formula is as follows: in , It is a positive linear proportionality coefficient. is a penalty constant, and Fitness is the fitness function evaluation value in the particle swarm optimization algorithm, so that the higher the degree of class overlap, the smaller the fitness evaluation value; When overlap is eliminated, that is... and When the fitness is calculated, the formula is: 。 3. The bearing fault diagnosis system according to claim 1, characterized in that, The offline parameter optimization unit and the online simulation inference unit adopt a communication mode of one-time configuration or periodic update; when there is no update instruction, the online simulation inference unit independently performs pure hardware diagnosis based on the configured parameters, without needing to maintain real-time communication with the host computer.

4. The bearing fault diagnosis system according to claim 1, characterized in that, The voltage comparator receives an amplified DC characteristic voltage at its signal input terminal and a corresponding decision threshold that is amplified proportionally at its reference terminal. The control terminal of the analog switch is connected to the output terminal of the voltage comparator, its signal input terminal is independently connected to the system's constant drive power supply, and its signal output terminal is connected to the analog spiking neuron circuit.

5. A bearing fault diagnosis method based on gated spiking neurons and dual-channel simulation, characterized in that, Includes the following steps: S1: Collect the real-time vibration acceleration signal of the bearing, perform mean zeroing preprocessing, and then divide it into segments according to the preset length to construct a sample dataset; S2: In the host computer, the particle swarm optimization algorithm is used to jointly optimize the filter parameters of the first channel and the second channel. The optimal center frequency, bandwidth and first decision threshold of the first channel are output with the goal of maximizing the distance between the outer ring fault and other state classes. The optimal center frequency, bandwidth and second decision threshold of the second channel are output with the goal of maximizing the distance between the normal state and other state classes. S3: The optimal center frequency and bandwidth are sent to the microcontroller at the edge. The microcontroller dynamically configures the programmable switched capacitor filter chip to physically filter the input vibration signal and remove background noise. S4: Input the filtered AC signal into the true RMS converter chip to convert it into a DC characteristic voltage, and then amplify it through the linear operational amplifier circuit to output the amplified DC characteristic voltage. S5: The amplified DC characteristic voltage is input to the voltage comparator and compared with the corresponding decision threshold after proportional amplification; if the DC characteristic voltage is greater than the decision threshold, the voltage comparator outputs a high level to drive the analog switch to turn on, and the independent constant drive power supply is released to the analog spiking neuron circuit; if the DC characteristic voltage is not greater than the decision threshold, the analog switch remains off; S6: When the simulated spiking neuron circuit receives a conducting driving power supply, the internal membrane capacitor depolarizes and emits a pulse sequence; when it does not receive a driving power supply, it remains silent. S7: Detects the pulse output status of the first and second channels. When there is no pulse in both channels, the output (0,0) indicates normal operation. When there is a pulse only in the second channel, the output (0,1) indicates an inner ring fault. When there is a pulse in both channels, the output (1,1) indicates an outer ring fault.

6. The bearing fault diagnosis method according to claim 5, characterized in that, The decision threshold mentioned in step S5 is amplified and mapped proportionally on the hardware side according to the same amplification factor as the DC characteristic voltage.

Images