An ai-driven industrial equipment failure analysis method and system

Abstract

The present application relates to the technical field of industrial equipment diagnosis, in particular, the present application relates to an AI driven industrial equipment fault analysis method and system, solve the problem of insufficient multi-source data fusion, weak working condition adaptability and inaccurate fault positioning in traditional technology, the present application synchronously collects high-frequency time series signals through vibration, temperature and other multi-sensor and marks working condition parameters, generates dynamic confidence weight based on working condition matching historical fault library, extracts cross-modal features through space-time convolution network after fusing multi-source data, decouples the fault contribution degree of mechanical transmission, power and heat dissipation system by using fault mode separation module, realizes accurate diagnosis by combining frequency domain fingerprint, time domain waveform analysis and cross decision verification, finally predicts the remaining life through long short-term memory network and weibull model, the system integrates multi-source perception, dynamic fusion and intelligent diagnosis unit, improves the fault warning accuracy and equipment maintenance scientificity under complex working conditions, and provides a solution for intelligent diagnosis of industrial equipment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial equipment diagnosis, in particular to an AI-driven industrial equipment fault analysis method and system. BACKGROUND

[0002] Industrial equipment diagnosis technology is an important technology. In modern industrial production, industrial equipment fault analysis technology is a core means for ensuring continuous operation of production lines and reducing maintenance costs by monitoring equipment operating status in real time and identifying potential fault hazards. With the increasing intelligence and complexity of equipment, multi-sensor data fusion analysis and early fault warning have become key requirements.

[0003] However, the traditional fault analysis technology has core problems of insufficient multi-source data fusion and weak working condition adaptability. The existing scheme independently analyzes single-sensor data such as vibration and temperature, and does not establish a correlation model between working condition parameters and sensor signals. When the load intensity or environmental temperature and humidity changes, the signal-to-noise ratio of sensor data fluctuates, causing feature extraction deviation, and the mechanical fault misdiagnosis rate increases. Fixed weight fusion of multi-source data changes the fault feature contribution degree of different sensors when the equipment speed changes, and the power system fault misses the rate increases. In addition, the traditional method lacks equipment level decoupling capability and cannot distinguish the fault contribution degree of mechanical transmission, power, and heat dissipation systems. When multiple systems have concurrent faults, the positioning accuracy is low, leading to increased maintenance blindness and prolonged unplanned downtime. This rough data fusion and lack of working condition adaptability ultimately result in low fault warning accuracy of the traditional scheme in complex industrial scenarios, high maintenance costs, and difficulty in meeting the reliability requirements of intelligent manufacturing. In order to solve this technical problem, we provide an AI-driven industrial equipment fault analysis method and system. SUMMARY

[0004] The purpose of the present application is to provide an AI-driven industrial equipment fault analysis method and system to solve the problems raised in the background.

[0005] 1. Since the traditional scheme has insufficient multi-source data fusion and feature extraction deviation when the working condition changes, the present case dynamically fuses multi-source data with confidence weight, matches historical fault library with working condition parameters, which can improve the data fusion accuracy in complex working conditions and reduce the mechanical fault misdiagnosis rate.

[0006] 2. Since the traditional method lacks equipment level decoupling capability and the positioning of multiple system concurrent faults is not accurate, the present case distinguishes the fault contribution degree of each subsystem through a decoupling neural network, which can realize accurate fault positioning, reduce maintenance blindness, and improve the fault positioning accuracy.

[0007] To achieve the above object, one of the objects of the present application is to provide an AI-driven industrial equipment fault analysis method, comprising the following steps:

[0008] S1, through the vibration sensor, temperature sensor, acoustic sensor and current sensor embedded in the industrial equipment, high-frequency time sequence signals are synchronously collected, and equipment working condition parameters are marked, the equipment working condition parameters including load intensity, environmental temperature and humidity and running speed;

[0009] S2, based on the working condition parameter matching historical fault database, the feature correlation coefficient of each sensor data and the typical fault mode under the current working condition is calculated, the real-time signal-to-noise ratio of each sensor signal is dynamically evaluated by using the transfer learning model, and the dynamic confidence weight of each data source is generated according to the feature correlation coefficient and the real-time signal-to-noise ratio;

[0010] S3, the weighted multi-source data is input into the space-time convolution network, the cross-modal fusion features are extracted, the fusion features are decoupled at the equipment level through the fault mode separation module, the fault contribution degree of the mechanical transmission system, the power system and the heat dissipation system is distinguished, and the fault positioning result with mechanism explainability and the remaining service life prediction are output.

[0011] As a further improvement of the technical solution, the acquisition of the dynamic confidence weight in S2 includes:

[0012] The load intensity, the environmental temperature and humidity and the running speed are input into the fully connected neural network, and a 128-dimensional working condition feature vector is output, the cosine similarity of the current working condition feature vector and the fault mode label in the historical fault database is calculated, the feature correlation coefficient is generated, the spectral kurtosis and envelope entropy of the sensor signal are extracted through the transfer learning model, and the real-time signal-to-noise ratio score is generated according to the spectral energy distribution stability, the feature correlation coefficient and the real-time signal-to-noise ratio score are normalized and weighted summed, and the dynamic confidence weight of each data source is output.

[0013] As a further improvement of the technical solution, the extraction of the cross-modal fusion features in S3 includes parallel extraction of space-time features and dynamic weight injection:

[0014] The space-time features include spatial dimensions and time dimensions, the spatial dimensions are obtained by capturing the spatial topological correlation of multi-sensor data by using a dilated convolution layer, and the convolution kernel span is the number of sensors, and the time dimensions are obtained by capturing the long-term dependence of high-frequency time sequence signals through a gated recurrent unit;

[0015] The dynamic confidence weight of each data source is converted into a channel attention coefficient, and the feature map channel response of the convolution layer is weighted and adjusted.

[0016] As a further improvement of the technical solution, it further includes frequency domain fingerprint extraction, time domain waveform analysis and cross decision verification:

[0017] The frequency domain fingerprint extraction: performing fast Fourier transform on the weighted vibration sensor signal, and extracting the energy proportion at the fundamental frequency harmonic frequency multiplication as a mechanical fault fingerprint;

[0018] The time domain waveform analysis: calculating the waveform distortion rate of the current sensor signal, and triggering the secondary verification mechanism of the frequency domain fingerprint when the distortion rate exceeds the threshold;

[0019] The cross-decision verification: when the time domain distortion rate and the frequency domain energy distribution conflict, automatically increasing the acoustic sensor confidence weight for arbitration.

[0020] As a further improvement of the technical solution, the device level decoupling in S3 includes:

[0021] A three-layer decoupling neural network is constructed, including an input layer, a hidden layer and an output layer. The input layer receives a cross-modal fusion feature vector. The hidden layer separates the feature subspaces of the mechanical transmission system, the power system and the heat dissipation system through a sparse autoencoder. The output layer calculates the fault activation intensity of each subsystem and outputs the fault contribution percentage.

[0022] As a further improvement of the technical solution, the mechanical transmission system fault diagnosis further includes:

[0023] If the harmonic energy growth of the vibration signal at 1-3 times the rotation frequency exceeds 80% of the normal rate, and the signal-to-noise ratio of the acoustic sensor in this frequency band is greater than or equal to 20dB, it is determined as bearing inner ring fault;

[0024] When the acoustic signal appears a modulated side frequency at the meshing frequency sideband, the gear wear grade is confirmed in combination with the temperature gradient change rate.

[0025] As a further improvement of the technical solution, the power system fault diagnosis further includes:

[0026] When the third harmonic component of the current sensor grows to more than 15% of the fundamental wave, and the temperature sensor monitors a temperature rise rate greater than 5℃ / min at the motor winding, an insulation aging warning is triggered;

[0027] If the current waveform appears a millisecond-level pulse drop, and the vibration sensor detects impact energy in the motor axial direction, it is located as a winding inter-turn short circuit.

[0028] As a further improvement of the technical solution, the heat dissipation system fault diagnosis further includes:

[0029] The turbulent sound pressure level of the 200-500Hz frequency band is separated from the acoustic signal, and the heat dissipation air volume attenuation curve is constructed. When the air volume attenuation rate is greater than 30% and the temperature sensor inlet / outlet temperature difference is reduced by 40%, the radiator blockage position is determined.

[0030] As a further improvement of the technical solution, the remaining service life prediction comprises:

[0031] The failure contribution degrees of the mechanical transmission system, the power system and the heat dissipation system are input into the long short-term memory network and the Weibull mixed model, the remaining life of each subsystem is calculated, the minimum value of the remaining life of the three subsystems is taken as the overall remaining life of the equipment, and the dominant failure subsystem is marked.

[0032] The second object of the present application is to provide a system for realizing an AI-driven industrial equipment failure analysis method, comprising:

[0033] The multi-source sensing unit integrates four types of industrial sensors of vibration, temperature, acoustics and current, carries the working condition parameter acquisition module to acquire equipment working condition parameters, and the equipment working condition parameters include load intensity, environmental temperature and humidity and running speed;

[0034] The dynamic fusion unit arranges the working condition feature mapping neural network and the transfer learning signal-to-noise ratio evaluation module, outputs the dynamic confidence weight matrix, and injects the space-time convolution network in real time;

[0035] The intelligent diagnosis unit contains a cross-modal feature fusion module, a fault decoupling engine and a life predictor, and outputs a fault positioning map with subsystem contribution degree and a remaining life confidence interval.

[0036] Compared with the prior art, the present application has the following advantages:

[0037] The present application synchronously collects high-frequency time sequence signals such as vibration and temperature by multi-source sensors and marks the working condition parameters, provides multi-dimensional data support for fault analysis, ensures the comprehensiveness of data and the working condition correlation, matches the historical fault library based on the working condition parameters, calculates the feature correlation coefficient and generates the dynamic confidence weight combined with the transfer learning model, realizes the adaptive fusion of multi-source data, improves the accuracy of data fusion under complex working conditions, reduces the feature extraction deviation caused by working condition changes, inputs the weighted data into the space-time convolution network to extract the cross-modal fusion features, performs equipment level decoupling through the fault mode separation module, can accurately distinguish the failure contribution degrees of mechanical transmission, power and heat dissipation systems, realizes accurate positioning of faults and mechanism explainability analysis, improves the reliability of fault diagnosis combined with frequency domain fingerprint extraction, time domain waveform analysis and cross decision verification mechanism, predicts the remaining service life through the long short-term memory network and the Weibull mixed model, provides a scientific basis for equipment maintenance, and ensures the safe operation of the equipment. BRIEF DESCRIPTION OF DRAWINGS

[0038] Fig. 1 The overall working process chart of the present application is shown in the figure;

[0039] Fig. 2Fig. 1 is a schematic diagram of the overall structure of the present application;

[0040] The meanings of various reference numerals in the figures are as follows:

[0041] 1, multi-source perception unit; 2, dynamic fusion unit; 3, intelligent diagnosis unit. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0043] Please refer to Figs. 1-2 One of the purposes of the present embodiment is to provide an AI-driven industrial equipment fault analysis method, which comprises the following steps:

[0044] S1, synchronously collecting high-frequency time sequence signals through vibration sensors, temperature sensors, acoustic sensors and current sensors embedded on the industrial equipment, and marking equipment working condition parameters, the equipment working condition parameters including load intensity, environmental temperature and humidity and running speed;

[0045] To realize accurate analysis of industrial equipment faults, first, high-frequency time sequence signals are synchronously collected through multiple types of sensors and working condition parameters are marked, and the specific implementation is as follows:

[0046] To obtain comprehensive equipment running state data, vibration sensors, temperature sensors, acoustic sensors and current sensors are embedded at key parts of the industrial equipment, and the four types of sensors synchronously collect vibration acceleration, temperature, sound pressure level and current signals through a unified clock source to provide original data consistent in time and space for subsequent cross-modal analysis. At the same time of collecting sensor signals, equipment working condition parameters including load intensity, environmental temperature and humidity and running speed are obtained in real time through an industrial bus. The system attaches working condition parameter labels to each set of sensor time sequence data. The labels are marked using hardware time stamp synchronization technology to accurately match the collection time of working condition parameters with sensor signal sampling points, ensuring that subsequent analysis can be based on specific working conditions to evaluate signal characteristics and provide accurate working condition input for dynamic confidence weight calculation, effectively solving the problems of incomplete data collection and missing working condition correlation in traditional schemes

[0047] S2, based on the working condition parameter matching historical fault database, calculating the feature correlation coefficient of each sensor data and the typical fault mode under the current working condition, dynamically evaluating the real-time signal-to-noise ratio of each sensor signal using a transfer learning model, and generating the dynamic confidence weight of each data source according to the feature correlation coefficient and the real-time signal-to-noise ratio;

[0048] To solve the problem of fixed fusion weight of multi-source data in traditional solutions and poor adaptability to working conditions, dynamic confidence weights need to be generated based on working condition parameters and sensor signal characteristics. The acquisition of dynamic confidence weights in S2 includes:

[0049] To map multi-dimensional working condition parameters into a calculable feature vector, load intensity, environmental temperature, environmental humidity, and operating speed are input into a three-layer fully connected neural network, with 4 neurons in the input layer, 64 neurons in the hidden layer, and 128 neurons in the output layer. The hidden layer uses a ReLU activation function, and the output layer is not activated. For example, at a certain moment, the working condition parameters are load 80%, temperature 35°C, humidity 50% RH, and speed 1800 rpm. After network calculation, the 128-dimensional working condition feature vector Vc = [v1, v2,..., v128] is output. This vector compresses the spatiotemporal correlation characteristics of the working condition parameters. To measure the correlation degree of the current working condition and historical faults, the cosine similarity of the working condition feature vector and each fault mode label vector in the historical fault database is calculated. To adapt to the signal quality fluctuations caused by changes in working conditions, a pre-trained transfer learning model is used to extract the spectral kurtosis and envelope entropy of the sensor signals, and the standard of spectral energy distribution is calculated as a stability index. A real-time signal-to-noise ratio score of 0-1 is generated through linear mapping. The feature correlation coefficient matrix C and the real-time signal-to-noise ratio score S are normalized and weighted summed. The weight coefficient α = 0.6, the working condition correlation weight β = 0.4, and the signal-to-noise ratio weight. The formula is W i = α · (c i -min(C) / max(C)-min(C))+ β · s i , where W i is the dynamic confidence weight of the i-th sensor, c i is the feature correlation coefficient matrix of the i-th sensor, s i is the real-time signal-to-noise ratio score of the i-th sensor. This weight increases the weight of high signal-to-noise ratio sensor signals in high correlation working conditions, effectively reducing the feature extraction deviation caused by changes in working conditions, and solving the misdiagnosis and missed diagnosis problems caused by fixed weights in traditional solutions.

[0050] S3, input the weighted multi-source data into a spatiotemporal convolutional network to extract cross-modal fusion features. Through a fault mode separation module, the fusion features are decoupled at the device level to distinguish the fault contribution of mechanical transmission systems, power systems, and cooling systems. The output is a fault location result with mechanism explainability and a remaining useful life prediction.

[0051] In order to extract deep cross-modal features from multi-source sensor data and improve the accuracy of fault diagnosis, the weighted multi-source data needs to be extracted in parallel in time and space, and dynamic weights need to be injected to adjust the feature response. The cross-modal fusion features extracted in S3 include parallel extraction of time and space features and injection of dynamic weights, and the specific implementation is as follows:

[0052] In order to obtain the spatial correlation features between multi-sensor data, the weighted sensor signal matrix is processed by using a cavity convolution layer. The dimension of the sensor signal matrix is the number of sampling points x the number of sensors, for example, 4 types of sensors such as vibration, temperature, acoustics and current, and the number of sampling points is 1024. Therefore, the matrix dimension is 1024x4. The convolution kernel span of the cavity convolution layer is set to the number of sensors, i.e. 4, and the cavity rate is set to 1. The convolution kernel weight is obtained by training and learning. For example, the dimension of the convolution kernel is [4x1x64], where 64 is the output channel number. The spatial correlation features of different sensor data at the same time are captured by performing convolution operation between the convolution kernel and the input matrix, such as the cooperative change mode of vibration and current signals. Compared with the traditional convolution layer, the cavity convolution layer expands the capture range of spatial features by 2 times without increasing the calculation amount, which is used to improve the extraction accuracy of multi-sensor spatial topology correlation. In order to process the time series features of high-frequency time series signals, the output features of the cavity convolution layer are processed by using a gated recurrent unit network. The gated recurrent unit contains an update gate and a reset gate, which captures the long-term dependence of the signal through the forgetting and memory mechanism. The input is the feature sequence output by the cavity convolution layer, i.e. the feature dimension x the number of sampling points. For example, the feature dimension is 64 and the number of sampling points is 1024. The hidden layer dimension of the gated recurrent unit is set to 128. The feature vector containing time dependence is calculated by forward propagation. For example, for the vibration signal sequence in the bearing fault development process, the gated recurrent unit can effectively remember the time correlation between the early weak impact features and the late violent vibration features. In order to strengthen the features corresponding to the high-confidence sensors, the dynamic confidence weight of each data source is converted into a channel attention coefficient. First, the dynamic confidence weight is normalized. For example, the weights of vibration, temperature, acoustics and current are [0.9, 0.7, 0.8, 0.6], and after normalization, they are [0.32, 0.25, 0.29, 0.14]. Then, it is expanded to the same dimension as the channel number of the convolution layer output feature map. The feature map response is adjusted by weighting operation in the channel dimension to obtain the weighted feature map. Through the above parallel extraction of time and space features and dynamic weight injection operations, the system realizes the cross-modal feature fusion of multi-source sensor data, providing high-discrimination feature representation for subsequent fault location and life prediction.

[0053] In order to further improve the reliability of fault diagnosis, after extracting the cross-modal fusion features, multi-dimensional fault feature verification needs to be performed in combination with frequency domain fingerprints, time domain waveforms and cross decision verification. The specific implementation is as follows:

[0054] Further including frequency domain fingerprint extraction, time domain waveform analysis and cross decision verification:

[0055] In order to capture the characteristic frequency component of mechanical failure, the weighted vibration sensor signal is subjected to fast Fourier transform to convert the time domain signal into frequency domain spectrum, the fundamental frequency f0=rotational speed / 60 is calculated according to the running speed of the equipment, and then the energy proportion at 50Hz, 75Hz, 100Hz and 125Hz, i.e. the energy proportion at 50Hz, 75Hz, 100Hz and 125Hz, is extracted as the mechanical failure fingerprint, and the energy proportion calculation formula is: Where X(f i ) is the spectral amplitude of the i-th harmonic, n is the number of spectral points, and the waveform distortion rate of the current sensor signal is calculated for monitoring the abnormality of the power system. First, the fundamental wave component is extracted by a sliding window, and then the root mean square error of the actual waveform and the fundamental wave component is calculated. The distortion rate formula is: distortion rate= root mean square error / fundamental wave amplitude x 100%, the preset distortion rate threshold is 10%, when the calculated value exceeds the threshold, such as the distortion rate is 15% at a certain time, the secondary verification mechanism of the frequency domain fingerprint is automatically triggered to ensure the accurate distinction of the failure type, when the current signal distortion rate and the vibration signal frequency energy distribution conflict, such as the distortion rate is greater than 10% but the frequency energy has no obvious abnormality, the system automatically enhances the confidence weight of the acoustic sensor for arbitration, the arbitration mechanism is based on the following rules: if the time domain distortion rate exceeds the threshold and the frequency energy change is less than 20%, the confidence weight of the acoustic sensor is increased from the initial value 0.8 to 0.95, and the sound pressure level feature in the frequency band of 200-500Hz is extracted for auxiliary judgment, for example, the current distortion rate is 18% but the vibration frequency energy is normal, after increasing the acoustic weight, it is found that the gear meshing frequency sideband energy increases by 15dB, and finally it is determined that the gear is slightly worn rather than a power failure, which effectively solves the limitation problem of single feature diagnosis, provides accurate diagnosis basis for rapid maintenance of industrial equipment, and effectively avoids non-planned shutdown caused by single feature misjudgment.

[0056] In order to realize the accurate positioning of industrial equipment failure and solve the problem of inaccurate positioning of multiple system concurrent failures in traditional methods, a three-layer decoupling neural network is needed to decouple the cross-modal fusion features at the equipment level, and the specific implementation is as follows:

[0057] The equipment level decoupling in S3 includes:

[0058] To distinguish the fault contribution degree of mechanical transmission, power and heat dissipation system, a three-layer decoupling neural network is constructed, including input layer, hidden layer and output layer. The input layer receives a 128-dimensional cross-modal fusion feature vector, which is extracted by a spatiotemporal convolution network and contains the features of multiple sensors associated with space and time. The number of input layer neurons is consistent with the feature dimension. The feature vector is transmitted to the hidden layer through full connection. The time delay of input processing is controlled within 10m to ensure real-time performance. The hidden layer separates the feature subspaces of mechanical transmission system, power system and heat dissipation system through a sparse autoencoder, which contains 256 neurons. Decoupling is achieved through the following steps: first, the input features are mapped to the latent space using an encoder, and the encoder weight matrix bias The encoding formula is h = σ(W e ·F + b e ), where σ is the ReLU activation function. Then, the input features are reconstructed by the decoder, and the decoder weight matrix bias The reconstruction formula is During training, a sparsity constraint term is introduced to ensure that only 10% of the hidden layer neurons are in an active state, thereby forcing the network to learn independent feature representations of each subsystem. The output layer contains 3 neurons corresponding to the fault activation intensity of the mechanical transmission system, power system and heat dissipation system, respectively. The hidden layer features h are mapped to the output layer through a fully connected layer, and the weight matrix bias The activation function is Softmax, and the calculation formula is a = Softmax(W o ·h + b o ), where a = [a1, a2, a3], and a i represents the fault activation intensity of the i-th subsystem. The final fault contribution percentage p i = a i × 100%, and the calculation of the output layer ensures that the quantization error of each subsystem's fault contribution degree is less than 5%, providing an interpretable quantitative basis for fault location and effectively solving the problems of fuzzy fault location and high maintenance blindness in traditional methods.

[0059] After completing the device-level decoupling, for the fault diagnosis of the mechanical transmission system, the vibration and acoustic signal features and temperature gradient change rate are combined to achieve accurate identification of bearing and gear faults. The specific implementation is as follows:

[0060] The mechanical transmission system fault diagnosis further includes:

[0061] To avoid single signal misjudgment, the frequency domain features of vibration signal and the signal-to-noise ratio of acoustic signal are combined to determine the bearing inner ring fault. First, the 1-3 times rotational frequency range is calculated according to the running speed of the equipment, for example, when the speed is 1800 rpm, the fundamental frequency is 30 Hz, and the 1-3 times rotational frequency is 30-90 Hz. Then the harmonic energy growth rate in this frequency band is calculated, and the formula is: Where E 当前 is the 1-3 times rotational frequency energy at the current time, E 基准 is the reference energy in the normal state of the equipment, and △t is the time interval. When the growth rate exceeds 80% of the normal rate, and the signal-to-noise ratio of the acoustic sensor in this frequency band is greater than or equal to 20 dB, the signal-to-noise ratio = signal energy / noise energy, it is determined that the bearing inner ring is faulty. To quantify the degree of gear wear, the modulation side frequency of the acoustic signal and the temperature gradient change rate are used to confirm the wear grade. First, the gear meshing frequency in the acoustic signal is extracted For example, when the speed is 1500 rpm and the number of teeth is 20, the meshing frequency is 500 Hz. Then detect whether the sideband exists modulation side frequency, i.e. f m ±n×rotational frequency, n=1, 2, 3, and calculate the temperature gradient change rate of the gearbox, the formula is Where T 当前 is the current temperature, and T 1小时前 is the temperature one hour ago. According to the number of modulation side frequencies and the temperature gradient change rate, the wear grade is confirmed: 1 modulation side frequency and change rate less than 1℃ / h for light wear, 2 side frequencies and change rate 1-3℃ / h for moderate wear, and 3 or more side frequencies and change rate >3℃ / h for severe wear. This provides an accurate grade division basis for gear maintenance.

[0062] After completing the decoupling of the equipment level, for the fault diagnosis of the power system, the current harmonic component, motor winding temperature and vibration impact characteristics need to be combined to realize accurate identification of insulation aging and winding inter-turn short circuit fault. The specific implementation is as follows:

[0063] The power system fault diagnosis further includes:

[0064] To early detect the hidden danger of motor insulation aging, the current three times harmonic component ratio and winding temperature rise rate are combined for early warning. First, the current sensor signal is subjected to fast Fourier transform to calculate the energy ratio of the three times harmonic component and the fundamental wave, and the formula is Where f0 is the current fundamental frequency determined by the grid frequency, X(f) is the spectral amplitude at frequency f, when the third harmonic proportion increases to more than 15% of the fundamental wave, and the temperature sensor at the motor winding monitors the temperature rise rate greater than 5℃ / min, the insulation aging warning is triggered to avoid the motor burnout accident caused by insulation failure, and to accurately locate the winding inter-turn short circuit fault, the current waveform pulse drop and motor axial vibration impact are combined to judge, first the current signal is analyzed by sliding window, the window length is 10ms, the step is 1ms, whether there is a millisecond level pulse drop, that is, the drop amplitude is greater than 20% of the fundamental wave amplitude, the duration is 1-5ms, for example, the normal current amplitude is 10A, a 2A pulse drop occurs and lasts for 3ms, at the same time, the vibration sensor collects acceleration signals in the axial direction of the motor, that is, parallel to the motor shaft, calculates the impact energy, which is represented by the product of the time domain signal peak value and the duration, unit: g·ms, when the impact energy is greater than 0.5g·ms, it is jointly located as a winding inter-turn short circuit. In this fault, the short circuit current causes the electromagnetic force to suddenly change, causing axial vibration impact, and the time difference between the two is less than 10ms, for example, at a certain time, the current waveform has a 2ms pulse drop, the amplitude drops from 8A to 6A, and at the same time, the axial vibration impact energy reaches 0.8g·ms, which is located to the 3rd turn of the winding. The accurate fault location information provides accurate fault location information for rapid repair of the motor, effectively reducing the unplanned downtime loss caused by power failure.

[0065] After completing the device level decoupling, for the fault diagnosis of the heat dissipation system, the turbulent sound pressure level of the acoustic signal, the heat dissipation air volume attenuation curve and the temperature sensor temperature difference change are combined to realize accurate determination of the heat dissipation blockage position. The specific implementation is as follows:

[0066] The heat dissipation system fault diagnosis further comprises:

[0067] In order to monitor the air volume change of the heat dissipation system, first separate the turbulent sound pressure level of the 200-500Hz frequency band from the acoustic sensor signal, which is positively correlated with the airflow turbulence intensity in the radiator. A band-pass filter is used to filter the signal, and the sound pressure level amplitude of the frequency band is calculated by root mean square. For example, under normal working conditions, the sound pressure level is 70dB. Then, based on historical data, a heat dissipation air volume attenuation curve is constructed, with sound pressure level as abscissa and actual air volume as ordinate. Through air volume sensor calibration, a mapping relationship Q=f(L p ) is established, where Q is the air volume, L p is the sound pressure level of 200-500Hz. When the radiator is gradually blocked, the airflow turbulence is enhanced, the sound pressure level is increased, the air volume is attenuated, and the curve slope is obtained by fitting 100 sets of calibration data, which can accurately reflect the dynamic relationship between air volume and sound pressure level. In order to accurately determine the blockage position of the radiator, the joint threshold of air volume attenuation rate and temperature sensor inlet / outlet temperature difference change rate is calculated, and the air volume attenuation rate formula is wherein Q 基准 is the reference air volume of the radiator cleaning state, Q 当前 is the real-time air volume, and the temperature sensor inlet / outlet temperature difference reduction rate formula is wherein ΔT 基准 is the reference temperature difference, such as the inlet / outlet temperature difference of 20℃ in the cleaning state, and ΔT 当前 is the real-time temperature difference, when the air volume attenuation rate is greater than 30% and the temperature difference reduction rate is greater than 40%, it is determined that the radiator is blocked, and the blocked position is located through the sound pressure level spatial distribution: if the sound pressure level on the left side of the radiator is 5dB higher than that on the right side, the blocked position is located in the left 1 / 3 area, which provides accurate blocked position information for the maintenance of the heat dissipation system, and effectively improves the equipment heat dissipation efficiency and operation reliability.

[0068] After completing the fault contribution degree analysis of each subsystem, in order to scientifically predict the remaining useful life of the equipment and guide the maintenance decision, the fault contribution degree needs to be input into the long short-term memory network and Weibull mixed model for remaining life prediction, and the specific implementation is as follows:

[0069] The remaining useful life prediction includes:

[0070] In order to capture the time series characteristics of fault development, the historical data of fault contribution degree of mechanical transmission, power and heat dissipation system are input into the long short-term memory network, which contains 2 layers of hidden layers, 128 neurons in each layer, the input dimension is 3, corresponding to the contribution degree of 3 subsystems, the output dimension is 1, that is, the remaining life prediction value, the mean square error loss function is used during training, the optimizer is adaptive moment estimation, and the learning rate is 0.001, for example, the fault contribution degree of a certain mechanical transmission system gradually increases from 10% to 30%, the long short-term memory network predicts its remaining life as 150 hours through the memory of historical growth trend, the prediction error of long short-term memory network for nonlinear degradation trend is less than 10%, compared with traditional linear extrapolation method, the prediction accuracy of accelerated degradation scene is improved, in order to quantify the uncertainty of remaining life prediction, Weibull mixed model is used to calculate the probability distribution of the remaining life of each subsystem, and the Weibull distribution probability density function is wherein λ is the characteristic life, k is the shape parameter, that is, dimensionless, λ and k are fitted based on the historical data of subsystem faults through maximum likelihood estimation, for the current fault contribution degree, the characteristic life is adjusted through the mapping function λ' = λ × (1-contribution degree), the mixed model generates the probability density curve of the remaining life by weighting multiple Weibull distributions, and the weight is determined by the fault mode proportion, which provides a quantitative risk evaluation basis for maintenance decision, and the overall remaining life of the equipment is the minimum value of the remaining life of the three subsystems, that is, 整体 L 机械 = min(L 电力 , L散热 ); wherein, L 机械 , L 电力 , L 散热 are respectively the remaining life prediction values of the three subsystems, and the dominant fault subsystem, i.e., the subsystem with the largest contribution degree, is marked, the determination method conforms to the physical law that the shortest life subsystem of the equipment determines the overall life, and effectively avoids the problems of excessive maintenance or maintenance lag, through the above-mentioned remaining life prediction method, the system realizes the intelligentization of the whole chain from fault diagnosis to life prediction, provides scientific quantitative support for predictive maintenance of industrial equipment, and effectively reduces the maintenance cost and unplanned downtime loss.

[0071] The second object of the present application is to provide a system for implementing the AI-driven industrial equipment fault analysis method of any one of the above.

[0072] The multi-source perception unit 1 integrates four types of industrial-grade sensors of vibration, temperature, acoustics and current, carries the working condition parameter acquisition module to acquire equipment working condition parameters, and the equipment working condition parameters include load intensity, environmental temperature and humidity and running speed;

[0073] The dynamic fusion unit 2 deploys a working condition feature mapping neural network and a transfer learning signal-to-noise ratio evaluation module, outputs a dynamic confidence weight matrix, and injects a space-time convolution network in real time;

[0074] The intelligent diagnosis unit 3 contains a cross-modal feature fusion module, a fault decoupling engine and a life predictor, and outputs a fault positioning map with subsystem contribution degree and a remaining life confidence interval.

[0075] The present application synchronously collects high-frequency time sequence signals through vibration, temperature and other multi-sensors and marks the working condition parameters, generates dynamic confidence weights based on the working condition matching historical fault library, extracts cross-modal features through the space-time convolution network after fusing multi-source data, decouples the fault contribution degree of mechanical transmission, power and heat dissipation systems by using the fault mode separation module, realizes accurate diagnosis by combining frequency domain fingerprint, time domain waveform analysis and cross decision verification, finally predicts the remaining life through the long short-term memory network and the Weibull model, improves the fault early warning accuracy and the scientific nature of equipment maintenance under complex working conditions, and provides a solution for intelligent diagnosis of industrial equipment.

[0076] The basic principles, main features and advantages of the present application are shown and described above. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, the above-mentioned embodiments and descriptions in the specification are only preferred examples of the present application, and are not intended to limit the present application, various changes and improvements can be made without departing from the spirit and scope of the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. An AI-driven method for industrial equipment fault analysis, characterized in that, Includes the following steps: S1. High-frequency timing signals are synchronously acquired by vibration sensors, temperature sensors, acoustic sensors and current sensors embedded in industrial equipment, and equipment operating parameters are marked. The equipment operating parameters include load intensity, ambient temperature and humidity and operating speed. S2. Based on the historical fault database matching the operating condition parameters, calculate the feature correlation coefficient between each sensor data and the typical fault mode under the current operating condition, use the transfer learning model to dynamically evaluate the real-time signal-to-noise ratio of each sensor signal, and generate the dynamic confidence weight of each data source according to the feature correlation coefficient and the real-time signal-to-noise ratio. S3. Input the weighted multi-source data into the spatiotemporal convolutional network, extract cross-modal fusion features, and decouple the fusion features at the device level through the fault mode separation module to distinguish the fault contribution of mechanical transmission system, power system and heat dissipation system, and output fault location results with mechanism interpretability and remaining service life prediction.

2. The AI-driven industrial equipment fault analysis method according to claim 1, characterized in that, The acquisition of dynamic confidence weights in S2 includes: Load intensity, ambient temperature and humidity, and operating speed are input into a fully connected neural network, which outputs a 128-dimensional operating condition feature vector. The cosine similarity between the current operating condition feature vector and the labels of each fault mode in the historical fault database is calculated to generate feature correlation coefficients. The spectral kurtosis and envelope entropy of the sensor signal are extracted through a transfer learning model, and a real-time signal-to-noise ratio score is generated based on the stability of the spectral energy distribution. The feature correlation coefficients and the real-time signal-to-noise ratio score are normalized and weighted, and the dynamic confidence weights of each data source are output.

3. The AI-driven industrial equipment fault analysis method according to claim 1, characterized in that, The extraction of cross-modal fusion features in S3 includes parallel extraction of spatiotemporal features and dynamic weight injection: The spatiotemporal features include spatial and temporal dimensions. The spatial dimension is obtained by capturing the spatial topological relationships of multi-sensor data using dilated convolutional layers, with the convolutional kernel span being the number of sensors. The temporal dimension is obtained by capturing the long-term dependencies of high-frequency time-series signals using gated recurrent units. The dynamic confidence weights of each data source are converted into channel attention coefficients, which are then used to weight and adjust the channel responses of the feature maps in the convolutional layers.

4. The AI-driven industrial equipment fault analysis method according to claim 3, characterized in that, Further steps include frequency domain fingerprint extraction, time domain waveform analysis, and cross-decision verification. The frequency domain fingerprint extraction involves performing a fast Fourier transform on the weighted vibration sensor signal to extract the energy proportion at the fundamental harmonic harmonics as a mechanical fault fingerprint. The time-domain waveform analysis involves calculating the waveform distortion rate of the current sensor signal. When the distortion rate exceeds a threshold, a secondary verification mechanism for the frequency-domain fingerprint is triggered. The cross-decision verification: when there is a conflict between the time domain distortion rate and the frequency domain energy distribution, the confidence weight of the acoustic sensor is automatically increased for arbitration.

5. The AI-driven industrial equipment fault analysis method according to claim 1, characterized in that, The device-level decoupling in S3 includes: A three-layer decoupled neural network is constructed, consisting of an input layer, a hidden layer, and an output layer. The input layer receives cross-modal fused feature vectors. The hidden layer separates the feature subspaces of the mechanical transmission system, the power system, and the heat dissipation system through a sparse autoencoder. The output layer calculates the fault activation intensity of each subsystem and outputs the percentage of fault contribution.

6. The AI-driven industrial equipment fault analysis method according to claim 5, characterized in that, The fault diagnosis of the mechanical transmission system further includes: If the harmonic energy of the vibration signal at 1-3 times the frequency increases by more than 80% of the normal rate, and the signal-to-noise ratio of the acoustic sensor in this frequency band is greater than or equal to 20dB, then it is determined to be a bearing inner ring fault. When the acoustic signal exhibits a modulation sideband in the meshing frequency sideband, the gear wear level is determined by combining this with the temperature gradient change rate.

7. The AI-driven industrial equipment fault analysis method according to claim 5, characterized in that, The fault diagnosis of the power system further includes: When the third harmonic component of the current sensor increases to more than 15% of the fundamental frequency, and the temperature sensor detects a temperature rise rate of more than 5°C / min at the motor winding, an insulation aging warning is triggered. If the current waveform shows a millisecond-level pulse drop, and the vibration sensor detects impact energy in the motor axis, it is located as an inter-turn short circuit in the winding.

8. The AI-driven industrial equipment fault analysis method according to claim 5, characterized in that, The fault diagnosis of the heat dissipation system further includes: Turbulent sound pressure levels in the 200–500 Hz frequency band are separated from the acoustic signal to construct a heat dissipation airflow attenuation curve. When the airflow attenuation rate is greater than 30% and the temperature difference between the inlet and outlet of the temperature sensor decreases by 40%, the location of the radiator blockage is determined.

9. The AI-driven industrial equipment fault analysis method according to claim 1, characterized in that, The remaining useful life prediction includes: The failure contributions of the mechanical transmission system, power system, and heat dissipation system are input into the Long Short-Term Memory Network and the Weibull hybrid model to calculate the remaining lifetime of each subsystem. The minimum value of the remaining lifetime of the three subsystems is taken as the overall remaining lifetime of the equipment, and the dominant failure subsystem is marked.

10. A system for implementing an AI-driven industrial equipment fault analysis method according to any one of claims 1-9, characterized in that, include: The multi-source sensing unit (1) integrates four types of industrial-grade sensors: vibration, temperature, acoustic and current, and is equipped with a working condition parameter acquisition module to collect equipment working condition parameters, including load intensity, ambient temperature and humidity and operating speed. The dynamic fusion unit (2) deploys a working condition feature mapping neural network and a transfer learning signal-to-noise ratio evaluation module, outputs a dynamic confidence weight matrix, and injects it into the spatiotemporal convolutional network in real time; The intelligent diagnostic unit (3) includes a cross-modal feature fusion module, a fault decoupling engine and a lifetime predictor, and outputs a fault location map with subsystem contribution and a remaining lifetime confidence interval.

Images