Servo driving system state monitoring method and system based on multi-sensor fusion

Abstract

The invention provides a servo driving system state monitoring method and system based on multi-sensor fusion, and the method comprises the steps: firstly obtaining a multi-source sensor data set, including vibration waveform, temperature gradient and current harmonic data, of target servo driving equipment, carrying out the time synchronization and amplitude calibration of the multi-source sensor data set, and generating a standardized monitoring data set; calling a dynamic coding model to extract operation characteristics to form an operation characteristic set, evaluating the operation characteristic set based on a multi-task state evaluation model to obtain a real-time health state index, an abnormal risk level and a performance degradation trend, and finally generating an equipment maintenance optimization strategy according to the results. Therefore, accurate state monitoring of the servo driving system is realized through multi-sensor fusion, and an effective decision basis can be provided for equipment maintenance.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and in particular to a servo drive system state monitoring method and system based on multi-sensor fusion. Background Art

[0002] In the field of industrial automation and intelligent manufacturing, the servo drive system is a core power component, and its operating status is directly related to the efficiency of the production line, product quality and equipment safety. With the advent of the Industrial 4.0 era, higher requirements are placed on the reliability and intelligent operation and maintenance level of the servo drive system. However, the existing servo drive system status monitoring technology still faces many challenges.

[0003] Related technologies often use static thresholds or simple statistical analysis methods, which are difficult to adapt to the dynamic changes in the equipment's operating status. Such methods cannot adaptively extract key features in equipment operation, and it is even more difficult to predict the degradation trend of equipment performance, resulting in delayed operation and maintenance decisions and increased risks of unplanned downtime. At the same time, due to the lack of a multi-task collaborative evaluation mechanism, existing technologies are difficult to simultaneously take into account the comprehensive analysis of equipment health status, abnormal risks, and performance degradation trends, resulting in a lack of a global perspective in the formulation of operation and maintenance strategies, making it difficult to achieve accurate and proactive maintenance operations. Summary of the invention

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a servo drive system state monitoring method based on multi-sensor fusion, the method comprising:

[0005] Acquire a multi-source sensor data set of a target servo drive device, wherein the multi-source sensor data set includes vibration waveform data, temperature gradient data, and current harmonic data;

[0006] Performing time synchronization and amplitude calibration processing on the multi-source sensor data set to generate a standardized monitoring data set;

[0007] Calling a dynamic coding model to extract operation features from the standardized monitoring data set to generate an operation feature set of a target device;

[0008] Performing a status assessment on the operation feature set based on a multi-task status assessment model to generate a real-time health status indicator, an abnormal risk level, and a performance degradation trend of the target device;

[0009] An equipment maintenance optimization strategy is generated according to the real-time health status indicators, abnormal risk levels and performance degradation trends, and the equipment maintenance optimization strategy is fed back to the operation and maintenance control platform to trigger active maintenance operations.

[0010] On the other hand, an embodiment of the present invention also provides a servo drive system status monitoring system based on multi-sensor fusion, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0011] Based on the above aspects, the embodiment of the present application realizes the global, deep perception and intelligent decision-making of the operating status of the target servo drive equipment, significantly improving the accuracy of equipment status monitoring and the intelligent level of operation and maintenance management. Specifically, by integrating multi-source heterogeneous sensor information such as vibration waveform data, temperature gradient data and current harmonic data, and through time synchronization and amplitude calibration processing, the spatiotemporal differences and dimensional barriers between multi-source data are eliminated, ensuring the consistency and accuracy of data fusion. The introduction of the dynamic coding model enables the operation feature extraction process to adaptively capture the dynamic changes of the equipment status and generate a highly distinguishable operation feature set, which significantly improves the sensitivity and specificity of the status assessment. Based on the above feature set, the multi-task status assessment model realizes the synchronous evaluation of the real-time health status indicators, abnormal risk levels and performance degradation trends of the equipment, which not only reveals the current operating status of the equipment, but also predicts potential failure risks and performance evolution trends. Ultimately, the equipment maintenance optimization strategy generated based on the evaluation results can accurately guide the operation and maintenance control platform to implement proactive maintenance operations, achieving closed-loop optimization from condition monitoring to operation and maintenance decision-making, effectively reducing equipment failure rates, and extending equipment service life, while improving production efficiency and operation and maintenance economy. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a schematic diagram of the execution flow of a servo drive system state monitoring method based on multi-sensor fusion provided in an embodiment of the present invention.

[0013] Figure 2 It is a schematic diagram of exemplary hardware and software components of a servo drive system state monitoring system based on multi-sensor fusion provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 The figure is a flow chart of a servo drive system state monitoring method based on multi-sensor fusion provided by an embodiment of the present invention. The servo drive system state monitoring method based on multi-sensor fusion is introduced in detail below.

[0015] Step S110: Acquire a multi-source sensor data set of a target servo drive device, wherein the multi-source sensor data set includes vibration waveform data, temperature gradient data, and current harmonic data.

[0016] In this embodiment, a servo drive device in a factory is taken as an example. The servo drive device is responsible for driving multiple mechanical parts to work together on the production line. In order to comprehensively monitor the operating status of the servo drive device, a variety of sensors are installed at key parts of the servo drive device. In detail, the vibration sensor is installed on the motor housing to collect vibration waveform data to reflect the vibration conditions during the operation of the servo drive device; the temperature sensor is arranged near the motor winding and on the surface of some key transmission parts to obtain temperature gradient data and understand the temperature changes of various parts of the servo drive device. The current sensor is installed on the power supply line of the servo drive device to collect current harmonic data in real time to analyze the electrical operation status of the servo drive device.

[0017] Assume that at a certain moment, the vibration waveform data collected by the vibration sensor presents a series of peaks and troughs, and the data is recorded as [V1, V2, V3, ..., Vn], where V1 represents the vibration amplitude at the first sampling moment, V2 represents the vibration amplitude at the second sampling moment, and so on, and n is the number of sampling points. The temperature gradient data collected by the temperature sensor, for example, the temperature sensor near the motor winding records the temperature value at regular time intervals, forming a temperature gradient data sequence [T1, T2, T3, ..., Tm], T1 is the temperature value at the starting moment, Tm is the temperature value at the mth sampling moment, and m is the number of temperature sampling points. The current harmonic data collected by the current sensor is also recorded in a sequence form, such as [I1, I2, I3, ..., Ik], I1 is the current value at the first sampling moment, Ik is the current value at the kth sampling moment, and k is the number of current sampling points.

[0018] Step S120: performing time synchronization and amplitude calibration processing on the multi-source sensor data set to generate a standardized monitoring data set.

[0019] In this embodiment, step S120 may include:

[0020] Step S121: extracting a first sampling timestamp sequence of the vibration waveform data, a second sampling timestamp sequence of the temperature gradient data, and a third sampling timestamp sequence of the current harmonic data.

[0021] In this embodiment, the vibration waveform data acquisition system can record the corresponding timestamp for each sampling point to form a first sampling timestamp sequence, such as [ts1, ts2, ts3, ..., tsn], where ts1 represents the timestamp of the first vibration data sampling point, and tsn represents the timestamp of the nth vibration data sampling point. The temperature gradient data acquisition system will also record the corresponding timestamp to form a second sampling timestamp sequence, such as [tt1, tt2, tt3, ..., ttm], where tt1 is the timestamp of the first temperature data sampling point, and ttm is the timestamp of the mth temperature data sampling point. The current harmonic data also has a corresponding third sampling timestamp sequence, such as [ti1, ti2, ti3, ..., tik], where ti1 is the timestamp of the first current data sampling point, and tik is the timestamp of the kth current data sampling point.

[0022] Step S122: determine the maximum common time interval between the first sampling timestamp sequence, the second sampling timestamp sequence and the third sampling timestamp sequence, perform time axis alignment processing on the vibration waveform data, temperature gradient data and current harmonic data based on a linear interpolation algorithm, and generate a time synchronization data set.

[0023] In this embodiment, after obtaining the first sampling timestamp sequence, the second sampling timestamp sequence and the third sampling timestamp sequence, it is necessary to find the maximum common time interval between them. Assume that after analysis, it is found that the maximum common time interval is Δt. For vibration waveform data, since its sampling frequency may be different from that of temperature gradient data and current harmonic data, for example, the sampling frequency of vibration waveform data is fs1, the sampling frequency of temperature gradient data is fs2, and the sampling frequency of current harmonic data is fs3. If fs1>fs2, then in the process of time axis alignment, for temperature gradient data, it is necessary to insert new data points between the original sampling points through a linear interpolation algorithm to align it with the vibration waveform data in time. For example, between the time points tsi and tsi+1 of the vibration waveform data, according to the linear interpolation algorithm, for the temperature gradient data, the interpolated temperature value of the corresponding time point in the time period is calculated. Similarly, similar processing is performed on the current harmonic data, and finally the vibration waveform data, temperature gradient data and current harmonic data are aligned on the time axis to form a time synchronization data set.

[0024] Step S123: Call a preset sensor calibration coefficient set to perform amplitude compensation processing on the vibration waveform data, temperature gradient data and current harmonic data in the time synchronization data set, wherein the sensor calibration coefficient set includes a frequency response compensation factor of the vibration sensor, a nonlinear error correction factor of the temperature sensor and a phase offset correction parameter of the current sensor.

[0025] In this embodiment, the preset sensor calibration coefficient set can be obtained through a large number of experiments and calibrations. For vibration waveform data, assuming that the frequency response compensation factor of the vibration sensor is Cf1, within a certain frequency band, due to the frequency response characteristics of the sensor, the collected vibration amplitude may have deviations. For example, under a vibration signal with a frequency of f1, the collected vibration amplitude is Va. According to the calibration coefficient Cf1, the vibration amplitude after amplitude compensation is Vb=Va×Cf1. For temperature gradient data, the temperature sensor may have a nonlinear error, and its nonlinear error correction factor is Ct. Assuming that the collected temperature value is Ta, considering the nonlinear error, the corrected temperature value Tb=Ta+Ct×(Ta-T0)², where T0 is the reference temperature value. For current harmonic data, the current sensor may have a phase offset, and its phase offset correction parameter is Ci. Assuming that the collected current value is Ia, the current value after phase offset correction Ib=Ia×exp(j×Ci), where the phase of the current is adjusted by the phase correction parameter to obtain a more accurate current amplitude. Through the above amplitude compensation processing operation, the amplitude of each sensor data can be made more accurate and reliable.

[0026] Step S124: normalizing the amplitude-compensated vibration waveform data, temperature gradient data, and current harmonic data to generate a standardized monitoring data set.

[0027] In this embodiment, after amplitude compensation, the corresponding data can be normalized. In detail, for the vibration waveform data, assuming that the vibration data sequence after amplitude compensation is [Vb1, Vb2, Vb3, ..., Vbn], its normalization formula is Vn_i=(Vb_i-min(Vb)) / (max(Vb)-min(Vb)), where Vn_i is the i-th vibration data after normalization, min(Vb) is the minimum value in the vibration data sequence after amplitude compensation, and max(Vb) is the maximum value. For the temperature gradient data, the temperature data sequence after amplitude compensation is [Tb1, Tb2, Tb3, ..., Tbm], the normalization formula is Tn_j=(Tb_j-min(Tb)) / (max(Tb)-min(Tb)), and Tn_j is the j-th temperature data after normalization. For current harmonic data, the current data sequence after amplitude compensation is [Ib1, Ib2, Ib3, ..., Ibk], and the normalization formula is In_k=(Ib_k-min(Ib)) / (max(Ib)-min(Ib)), where In_k is the kth current data after normalization. Through the above normalization process, the amplitude range of each sensor data is unified to [0, 1] to generate a standardized monitoring data set.

[0028] Step S130: calling the dynamic coding model to extract the operation characteristics of the standardized monitoring data set, and generating an operation characteristic set of the target device.

[0029] In this embodiment, step S130 may include:

[0030] Step S131: extracting time domain features from the vibration waveform data in the standardized monitoring data set to generate vibration effective value, peak factor and kurtosis coefficient.

[0031] For vibration waveform data, take the standardized vibration data sequence [Vn1, Vn2, Vn3, …, Vnn] as an example. The calculation method of the effective value of vibration is to first square the amplitude of each sampling point, then calculate the average of the above square values, and finally take the square root. That is, the effective value of vibration Veff=√((Vn1²+Vn2²+Vn3²+…+Vnn²) / n). The peak factor is the ratio of the vibration peak value to the effective value. First find the maximum value Vmax in the vibration data sequence, then the peak factor CF=Vmax / Veff. The kurtosis coefficient is used to measure the impact characteristics of the vibration signal. The calculation method is to first subtract the average value from the amplitude of each sampling point and then perform a fourth power operation, then calculate the average of the above fourth power value, and then divide it by the fourth power of the effective value. Assuming that the average value of the vibration data is Vavg, the kurtosis coefficient Kurtosis=(((Vn1-Vavg) 4 +(Vn2-Vavg) 4 +(Vn3-Vavg) 4 +…+(Vnn-Vavg) 4 ) / n) / Veff 4 .

[0032] Step S132: performing wavelet packet decomposition processing on the vibration waveform data to extract preset frequency band energy proportion characteristics and frequency band entropy value characteristics.

[0033] Step S1321: Select the Daubechies wavelet basis function to perform 5-layer wavelet packet decomposition on the vibration waveform data to obtain 32 frequency band node coefficients.

[0034] In this embodiment, the Daubechies wavelet basis function is used to decompose the vibration waveform data. The standardized vibration waveform data is decomposed by a five-layer wavelet packet. According to the principle of wavelet packet decomposition, 2 5 = 32 frequency band node coefficients. For example, the frequency band node coefficients obtained after decomposition are [W1, W2, W3, ..., W32], and each frequency band node coefficient represents the signal characteristics in different frequency bands.

[0035] Step S1322: Calculate the energy value of each frequency band node, and divide the energy value of each frequency band by the total energy to obtain a preset frequency band energy ratio feature.

[0036] For each frequency band node coefficient Wi, its energy value Ei=Wi². Total energy E=E1+E2+E3+…+E32. The preset frequency band energy proportion feature Pi=Ei / E, for example, P1=E1 / E, P2=E2 / E, and so on, thereby obtaining the energy proportion of each frequency band.

[0037] Step S1323: Calculate the permutation entropy of the coefficient sequence of each frequency band node to obtain the frequency band entropy value feature.

[0038] In this embodiment, for the coefficient sequence of each frequency band node, for example, the coefficient sequence of the i-th frequency band node [Wi1, Wi2, Wi3, …, Win], its permutation entropy is calculated. First, the embedding dimension m and the delay time τ are determined, assuming m=3, τ=1. The coefficient sequence is reconstructed into an m-dimensional vector. For example, for the frequency band node i=1, the reconstructed vector is [(W11, W12, W13), (W12, W13, W14), …]. Then the above vectors are sorted, the number of occurrences of different permutation orders is counted, and the permutation entropy of the frequency band node is calculated according to the permutation entropy calculation formula, denoted as Si. Such calculations are performed on all 32 frequency band nodes to obtain the frequency band entropy value characteristics.

[0039] Step S1324: filter the characteristic frequency bands whose energy proportion exceeds the set threshold, and add the corresponding frequency band entropy value characteristics and energy proportion characteristics to the operation feature set.

[0040] In this embodiment, assuming that the energy proportion threshold is set to Th, the energy proportion features of the calculated 32 frequency bands are screened. For example, if it is found that the energy proportions P5, P8, P12 of the 5th, 8th, 12th, etc. frequency bands exceed the threshold Th, then the frequency band entropy value features S5, S8, S12 and the energy proportion features P5, P8, P12 corresponding to the above frequency bands are added to the operation feature set.

[0041] Step S133: performing sliding window mean processing on the temperature gradient data to generate temperature change rate characteristics and heat accumulation trend characteristics.

[0042] In this embodiment, for the standardized temperature gradient data sequence [Tn1, Tn2, Tn3, ..., Tnm], sliding window mean processing is adopted. Assume that the sliding window size is w, for example, w=5. For the temperature change rate feature, taking the i-th window as an example, the data in the window is [Tn i , Tn i+1 , Tn i+2 , Tn i+3 , Tn i+4], temperature change rate characteristic Rt_i=(Tn i+4 -Tn i ) / 4. By moving the window, a series of temperature change rate characteristic values ​​[Rt1, Rt2, Rt3, ...] are obtained. For the heat accumulation trend characteristic, the sum of the temperature values ​​in each window is calculated, for example, the heat accumulation value Ht_i of the i-th window = Tn i +Tn i+1 +Tn i+2 +Tn i+3 +Tn i+4 , by moving the window, we get the heat accumulation trend characteristic sequence [Ht1, Ht2, Ht3, …].

[0043] Step S134: Perform fast Fourier transform processing on the current harmonic data to extract the fundamental wave amplitude, harmonic distortion rate and characteristic subharmonic component amplitude ratio.

[0044] In this embodiment, the standardized current harmonic data sequence [In1, In2, In3, ..., Ink] is subjected to fast Fourier transform processing, and the frequency domain data is obtained after fast Fourier transform. The amplitude corresponding to the fundamental frequency is found in the frequency domain, which is recorded as Ifundamental. The calculation method of the harmonic distortion rate THD is to first calculate the sum of the squares of the amplitudes of all harmonics except the fundamental wave, and then divide it by the square of the fundamental wave amplitude, and then multiply it by 100% after taking the square root. That is, THD=√((Ih1²+Ih2²+Ih3²+...) / Ifundamental²)×100%, where Ih1, Ih2, Ih3, etc. are the amplitudes of each harmonic. For the amplitude ratio of characteristic subharmonic components, assuming that the third and fifth harmonics are concerned, the amplitude ratio of characteristic subharmonic components R=Ih3 / Ih5.

[0045] Step S135: Input the vibration effective value, peak factor, kurtosis coefficient, preset frequency band energy proportion characteristics, frequency band entropy value characteristics, temperature change rate characteristics, heat accumulation trend characteristics, fundamental wave amplitude, harmonic distortion rate and characteristic subharmonic component amplitude ratio into the dynamic coding model, perform dynamic association coding through a multi-layer cross-attention mechanism, and generate an operation feature set.

[0046] In this embodiment, all the features extracted previously, namely, the effective value of vibration Veff, the peak factor CF, the kurtosis coefficient Kurtosis, the preset frequency band energy proportion feature [P5, P8, P12], the frequency band entropy value feature [S5, S8, S12], the temperature change rate feature [Rt1, Rt2, Rt3], the heat accumulation trend feature [Ht1, Ht2, Ht3], the fundamental amplitude Ifundamental, the harmonic distortion rate THD, and the characteristic subharmonic component amplitude ratio R can be input into the dynamic coding model. The dynamic coding model dynamically associates and encodes the above-mentioned different types of features through a multi-layer cross-attention mechanism. For example, in the first layer of the attention mechanism, the model analyzes the association between the vibration feature and the temperature feature, and performs weighted fusion according to their potential relationship in the operation of the device. In the second layer of the attention mechanism, the connection between the electrical feature (such as the current harmonic feature) and the previously fused feature is further considered, and weighted fusion and encoding are performed again. Finally, an operation feature set of the target device is generated, which contains multi-dimensional feature information after comprehensive analysis and encoding.

[0047] Step S140: performing a status assessment on the operation feature set based on a multi-task status assessment model to generate a real-time health status indicator, an abnormal risk level, and a performance degradation trend of the target device.

[0048] In this embodiment, step S140 may include:

[0049] Step S141: input the running feature set into the shared feature extraction layer to generate a shared latent feature vector.

[0050] In detail, the operation feature set contains feature information of multiple dimensions, which is input into the shared feature extraction layer of the multi-task state assessment model. The shared feature extraction layer is a convolutional neural network structure that can extract and compress the input operation feature set. For example, the operation feature set can be understood as a multi-dimensional feature matrix. The shared feature extraction layer performs sliding convolution operations on the matrix through convolution kernels to extract the key feature information. After operations such as convolution and pooling, the high-dimensional operation feature set is compressed into a low-dimensional shared implicit feature vector, which contains the key information in the operation feature set.

[0051] Step S142: inputting the shared implicit feature vector into the real-time health status prediction branch, the abnormal risk classification branch and the performance degradation regression branch respectively.

[0052] In this embodiment, after the shared implicit feature vector is generated, it is input into three different task branches respectively. Specifically, the real-time health status prediction branch is responsible for predicting the current health status of the device; the abnormal risk classification branch is used to determine the abnormal risk level currently faced by the device; and the performance degradation regression branch quantitatively evaluates the performance degradation trend of the device.

[0053] Step S143: In the real-time health status prediction branch, the shared implicit feature vector is subjected to temporal dependency modeling based on a gated recurrent unit network, and a real-time health status indicator is output.

[0054] In this embodiment, in the real-time health status prediction branch, a gated recurrent unit network (GRU) is used to process the shared implicit feature vector. The GRU network can capture the temporal dependencies in time series data. Assuming that the shared implicit feature vector is H, the GRU network will update the hidden state at the current moment according to the input at the current moment and the hidden state at the previous moment. For example, at time t, the input of the GRU network is H_t, and the hidden state at the previous moment is h_t-1. The hidden state h_t at the current moment is updated through the calculation formula of the GRU network. After processing for multiple time steps, a real-time health status indicator is finally output. The real-time health status indicator can be a numerical value, for example, between 0 and 100, representing the current health status of the device. The higher the numerical value, the better the health status.

[0055] Step S144: In the abnormal risk classification branch, the shared implicit feature vector is focused on key features based on the multi-head self-attention mechanism, and the probability distribution of the abnormal risk level is output.

[0056] In this embodiment, a multi-head self-attention mechanism is used in the abnormal risk classification branch. The multi-head self-attention mechanism can simultaneously focus on different parts of the shared implicit feature vector to focus on key features. For the shared implicit feature vector H, the multi-head self-attention mechanism will project it into multiple subspaces, for example, into 8 heads. Each head will calculate the attention weights between feature vectors and focus on key features through the above weights. For example, one head may pay more attention to features related to vibration anomalies, and another head may pay more attention to features related to temperature anomalies. After being processed by the multi-head self-attention mechanism, a probability distribution vector is output, each element of which represents the probability that the device is at a different abnormal risk level, for example, the probability distribution vector is [P1, P2, P3, P4], P1 represents the probability that the device is at a low risk level, P2 represents the probability of being at a medium risk level, P3 represents the probability of being at a high risk level, and P4 represents the probability of being at an extremely high risk level.

[0057] Step S145: In the performance degradation regression branch, nonlinear mapping is performed on the shared implicit feature vector based on a radial basis function network, and a quantized value of the performance degradation trend is output.

[0058] In this embodiment, the performance degradation regression branch uses a radial basis function network (RBF network) to process the shared implicit feature vector. The RBF network can perform nonlinear mapping on the input feature vector. For the shared implicit feature vector H, the RBF network uses a radial basis function as an activation function, such as a commonly used Gaussian radial basis function. Assuming that the input shared implicit feature vector H is a multidimensional vector [x1, x2, x3, ..., xn], the RBF network calculates the distance between the vector and each center vector ci, for example, the distance di=||H-ci|| (here ||.|| represents a certain norm, such as the Euclidean norm). Then, through a Gaussian radial basis function, such as exp(-di² / (2σ²)), where σ is a width parameter, the distance is mapped to an activation value. The RBF network performs a nonlinear transformation on the input through the above activation value, and finally outputs a quantized value of the performance degradation trend through linear combination. The quantized value can indicate the degree of degradation of device performance over time. For example, if the quantized value is positive, it indicates that the performance is gradually degrading. The larger the quantized value, the faster the degradation rate. If the quantized value is negative or close to 0, it indicates that the performance is relatively stable or has an improving trend.

[0059] Step S146: dynamically adjust the weight allocation ratio of the shared feature extraction layer according to the output results of the real-time health status prediction branch, the abnormal risk classification branch and the performance degradation regression branch to achieve multi-task collaborative optimization.

[0060] In this embodiment, step S146 may include:

[0061] Step S1461: Calculate the prediction error rate of the real-time health status prediction branch, the cross entropy loss value of the abnormal risk classification branch, and the mean square error of the performance degradation regression branch.

[0062] For the real-time health status prediction branch, assume that the real-time health status indicator it outputs is y_pred, and the actual health status label is y_true. The prediction error rate is calculated by first calculating the absolute error between the predicted value and the true value, and then finding the ratio of the average of the above errors to the average of the true value. That is, the prediction error rate E1=mean(|y_pred-y_true|) / mean(y_true). For the abnormal risk classification branch, its output is a probability distribution vector P_pred, and the actual abnormal risk level label is a one-hot encoding vector P_true (for example, if the device is at a medium risk level, the element corresponding to the medium risk level in P_true is 1, and the other elements are 0), and the cross entropy loss value E2=-Σ(P_true[i]*log(P_pred[i])), where i traverses all possible risk level categories. For the performance degradation regression branch, assume that the performance degradation trend quantization value it outputs is z_pred, and the actual performance degradation trend label is z_true, and the mean square error E3=mean((z_pred-z_true)²).

[0063] Step S1462: normalize the prediction error rate, cross entropy loss value and mean square error to obtain a weight adjustment factor for each branch.

[0064] In order to make different error indicators comparable, the calculated prediction error rate E1, cross entropy loss value E2 and mean square error E3 need to be normalized. Assume that the normalized prediction error rate is E1_norm=(E1-min(E1)) / (max(E1)-min(E1)), the cross entropy loss value is normalized to E2_norm=(E2-min(E2)) / (max(E2)-min(E2)), and the mean square error is normalized to E3_norm=(E3-min(E3)) / (max(E3)-min(E3)). The above normalized error values ​​are the weight adjustment factors of each branch, which are recorded as w1=E1_norm, w2=E2_norm, and w3=E3_norm respectively.

[0065] Step S1463: Dynamically update the channel attention weight of each convolution kernel in the shared feature extraction layer according to the weight adjustment factor.

[0066] In this embodiment, there are multiple convolution kernels in the shared feature extraction layer, and each convolution kernel is responsible for extracting different feature information. According to the weight adjustment factor, the channel attention weight of the convolution kernel is dynamically updated. For example, for the jth channel of a convolution kernel, its original attention weight is a_j, and the updated weight a_j'=a_j*(w1+w2+w3) / 3 (this is just a simple weighted update method, and more complex adjustments may be made according to the specific algorithm). In this way, the shared feature extraction layer can dynamically adjust the degree of attention to different features according to the performance of each branch task, thereby achieving multi-task collaborative optimization.

[0067] Step S1464: When the weight adjustment factor of the abnormal risk classification branch exceeds the preset alarm threshold, the parameter update of the performance degradation regression branch is frozen, and the abnormal risk classification task is optimized first.

[0068] In this embodiment, it is assumed that the preset alarm threshold is Threshold. When the weight adjustment factor w2 of the abnormal risk classification branch exceeds Threshold, it means that the error of the abnormal risk classification task is large and needs to be optimized first. At this time, the parameter update of the performance degradation regression branch is frozen, that is, its network parameters are no longer adjusted according to the error of the performance degradation regression branch. In this way, computing resources and optimization directions can be concentrated, the accuracy of abnormal risk classification can be improved first, and abnormal conditions of the equipment can be discovered and handled in a timely manner.

[0069] Step S150: Generate an equipment maintenance optimization strategy based on the real-time health status indicator, abnormal risk level and performance degradation trend, and feed back the equipment maintenance optimization strategy to the operation and maintenance control platform to trigger active maintenance operations.

[0070] In this embodiment, step S150 may include;

[0071] Step S151: Establish health status indicator threshold interval, abnormal risk level threshold interval and performance degradation trend slope threshold.

[0072] In this embodiment, through a large number of experiments and statistical analysis of equipment operation data, a corresponding threshold range is established. The threshold range of the health status index is set as [Health_min, Health_max]. For example, Health_min = 60 and Health_max = 80, indicating that when the real-time health status index is between 60 and 80, the equipment is in a relatively normal health state. The threshold range of the abnormal risk level is divided into thresholds of different levels. For example, the low-risk level threshold is Risk_low = 0.3, the medium-risk level threshold is Risk_mid = 0.6, and the high-risk level threshold is Risk_high = 0.8. The threshold of the performance degradation trend slope is set as Slope_threshold = 0.05, indicating that when the slope of the performance degradation trend exceeds 0.05, the performance degradation of the equipment needs to be concerned about.

[0073] Step S152: When the real-time health status index is lower than the lower limit of the health status index threshold range and the abnormal risk level exceeds the first risk threshold, generate an immediate shutdown and maintenance instruction.

[0074] In this embodiment, assume that the real-time health status index is Health_index and the abnormal risk level is Risk_level. When Health_index < Health_min and Risk_level > Risk_low, it indicates that the health condition of the equipment is poor and it faces a relatively high abnormal risk. For example, Health_index = 50 and Risk_level = 0.4. At this time, the system will generate an immediate shutdown and maintenance instruction to avoid further damage to the equipment and ensure production safety.

[0075] Step S153: When the slope of the performance degradation trend exceeds the slope threshold and the abnormal risk level is within the second risk threshold range, generate a preventive maintenance plan and spare part replacement suggestions.

[0076] In this embodiment, assume that the slope of the performance degradation trend is Slope. When Slope > Slope_threshold and Risk_low <= Risk_level <= Risk_mid, for example, Slope = 0.06 and Risk_level = 0.5, at this time, a preventive maintenance plan and spare part replacement suggestions need to be generated.

[0077] Step S1531: Calculate the predicted remaining service life according to the slope of the performance degradation trend.

[0078] In this embodiment, the remaining service life prediction value of the equipment can be calculated based on the performance degradation trend slope and the historical operation data and performance model of the equipment. For example, by establishing a linear model, assuming that the initial performance of the equipment is P0, the current performance is P1, the performance degradation trend slope is Slope, and the rated service life of the equipment is T0. Then the remaining service life prediction value Remaining_life=(P0-P1) / Slope. Assuming P0=100 (indicating the initial full performance state of the equipment), P1=80, Slope=0.06, then Remaining_life=(100-80) / 0.06≈333 (the unit can be hours, days, etc., here it is assumed to be hours).

[0079] Step S1532: query the equipment maintenance knowledge graph to match the historical maintenance case set corresponding to the remaining useful life prediction value.

[0080] In this embodiment, a large number of historical equipment maintenance cases are stored in the equipment maintenance knowledge graph, and query matching is performed in the equipment maintenance knowledge graph based on the calculated remaining service life prediction value. For example, a collection of historical equipment maintenance cases with a remaining service life between 300 and 350 hours is found. The above equipment maintenance cases contain maintenance measures and experiences of different equipment under similar performance degradation conditions.

[0081] Step S1533: extract the optimal maintenance time window, spare part model replacement records and labor cost data from historical maintenance cases.

[0082] In this embodiment, the optimal maintenance time window, spare part model replacement record and labor cost data of each case can be extracted from the matched historical maintenance case set. For example, in a historical case, the optimal maintenance time window is 50-80 hours before the remaining service life, the spare part model replacement record is the replacement of the motor brush and a bearing, and the labor cost is 5,000 yuan.

[0083] Step S1534: Based on the fuzzy decision algorithm, multi-objective optimization is performed on the maintenance time window, spare parts inventory status and production plan to generate the time nodes of the preventive maintenance plan, the required spare parts list and the human resource allocation plan.

[0084] In this embodiment, step S1534 may include:

[0085] Step S15341: convert the maintenance time window into a time overlap conflict matrix, map the spare parts inventory status into a spare parts availability vector, and parse the production plan into a maintenance operation time constraint sequence.

[0086] In this embodiment, for the maintenance time window, it is assumed that the multiple maintenance time windows obtained in the historical cases are [Time_window1, Time_window2, ...], which are converted into a time overlap conflict matrix. For example, Time_window1=[t1, t2], Time_window2=[t3, t4]. By analyzing whether there is overlap between the above time windows, a matrix is ​​constructed, and the elements in the matrix represent the overlap between different time windows. For the spare parts inventory status, it is assumed that there are multiple spare parts in the inventory, namely Spare_part1, Spare_part2, ..., which are mapped into spare parts availability vectors, for example, [Available1, Available2, ...], Available1 represents the available quantity of Spare_part1. For the production plan, the maintenance operation time constraint sequence is parsed. For example, the production plan requires that the equipment cannot be shut down for maintenance within a certain time period, and the above constraints are organized into a sequence.

[0087] Step S15342: construct a multi-objective optimization space based on the time overlap conflict matrix, spare parts availability vector and maintenance operation time constraint sequence, and use the Pareto frontier search algorithm to generate a set of candidate maintenance solutions.

[0088] In this embodiment, a multi-objective optimization space can be constructed using the time overlap conflict matrix, spare parts availability vector and maintenance operation time constraint sequence. The objectives in the multi-objective optimization space include minimizing the conflict between the maintenance time window and the production plan, maximizing the utilization of spare parts and minimizing labor costs, etc. A Pareto frontier search algorithm is used to search for a set of candidate maintenance solutions in the multi-objective optimization space. The candidate maintenance solutions achieve a balance between different objectives, and there is no situation where one solution is better than other solutions in all objectives.

[0089] Step S15343: performing time conflict resolution processing on the candidate maintenance solution set, eliminating candidate solutions that conflict with the maintenance operation time constraint sequence, and generating a conflict-free candidate maintenance solution set.

[0090] In this embodiment, for each of the candidate maintenance schemes, it is checked whether its maintenance time conflicts with the maintenance operation time constraint sequence. For example, if the maintenance time of a candidate scheme overlaps with the non-downtime time period specified in the production plan, the scheme is eliminated. After such processing, a set of conflict-free candidate maintenance schemes is generated.

[0091] Step S15344: sort the set of conflict-free candidate maintenance solutions by spare parts matching degree according to the spare parts availability vector, and select the top k candidate solutions with the highest spare parts matching degree.

[0092] In this embodiment, the matching degree between the spare parts required in each conflict-free candidate maintenance plan and the in-stock spare parts can be calculated according to the spare part availability vector. For example, if a certain candidate plan requires 5 Spare_part1, and the available quantity of Spare_part1 in stock is 8, then the matching degree of this plan for Spare_part1 is relatively high. By sorting the spare part matching degrees of all conflict-free candidate maintenance plans, the top k candidate plans with the highest spare part matching degrees are selected.

[0093] Step S15345: Input the top k candidate plans into the fuzzy decision-making device, and generate the time nodes of the preventive maintenance plan, the list of required spare parts, and the human resource allocation plan according to the preset maintenance priority rules.

[0094] In this embodiment, the selected top k candidate plans can be input into the fuzzy decision-making device. The fuzzy decision-making device comprehensively evaluates and makes decisions on the above plans according to the preset maintenance priority rules, such as giving priority to the continuity of the production plan and minimizing costs. Finally, the time nodes of the preventive maintenance plan are generated, for example, it is determined to perform maintenance 60 hours before the remaining service life; the list of required spare parts, such as listing spare parts like motor brushes and bearings that need to be replaced; and the human resource allocation plan, such as arranging several technicians to participate in the maintenance work, etc.

[0095] Step S154: When the real-time health status indicator is located in the middle of the health status indicator threshold range and the abnormal risk level is lower than the third risk threshold, generate an operation parameter optimization and adjustment plan.

[0096] In this embodiment, when Health_min < Health_index < Health_max and Risk_level < Risk_mid, for example, Health_index = 70 and Risk_level = 0.2, it means that the current health status of the equipment is relatively good and the abnormal risk is low. At this time, an operation parameter optimization and adjustment plan is generated. By adjusting some operation parameters of the equipment, such as the rotation speed and voltage of the motor, the operation efficiency and stability of the equipment are further improved. For example, according to the operation characteristics and historical data of the equipment, it is found that when the motor rotation speed is adjusted from the current 1500 revolutions per minute to 1450 revolutions per minute, the energy consumption of the equipment can be reduced by 5% without affecting the production quality. Then the operation parameter optimization and adjustment plan may include measures such as adjusting the motor rotation speed to 1450 revolutions per minute and the corresponding voltage adjustment.

[0097] Step S155: Integrate the immediate shutdown for repair instruction, the preventive maintenance plan, the spare part replacement suggestion, and the operation parameter optimization and adjustment plan into an equipment maintenance optimization strategy, and feedback the equipment maintenance optimization strategy to the operation and maintenance control platform to trigger proactive maintenance operations.

[0098] In this embodiment, the generation of immediate shutdown and maintenance instructions, preventive maintenance plans, spare parts replacement suggestions, and operating parameter optimization and adjustment plans can be integrated together to form a complete equipment maintenance optimization strategy. The strategy is then fed back to the operation and maintenance control platform. After receiving the strategy, the operation and maintenance control platform will trigger the corresponding active maintenance operations according to the content of the strategy. For example, if it is an immediate shutdown and maintenance instruction, the operation and maintenance control platform will send an instruction to the equipment control system to stop the equipment immediately and notify the maintenance personnel to perform maintenance; if it is a preventive maintenance plan and spare parts replacement suggestion, the operation and maintenance control platform will arrange for maintenance personnel to perform maintenance work at the specified time and prepare the required spare parts; if it is an operating parameter optimization and adjustment plan, the operation and maintenance control platform will send the instruction to adjust the parameters to the controller of the equipment to adjust the equipment operating parameters.

[0099] For example, in a further embodiment, the method may further include a training step of a multi-task state evaluation model:

[0100] Step S210: Acquire a multi-task training data set, wherein the multi-task training data set includes a historical operation feature set and its corresponding health status labels, abnormal risk level labels, and performance degradation trend labels.

[0101] For example, step S210 may include:

[0102] Step S2101: Collect multi-source sensor monitoring data of the servo drive device during its entire life cycle, perform time synchronization and amplitude calibration processing to generate a standardized historical monitoring data set.

[0103] In detail, during the entire life cycle of the servo drive equipment, its multi-source sensor monitoring data is continuously collected. For example, starting from the equipment installation and commissioning stage, vibration sensors, temperature sensors, and current sensors continuously collect data. At different stages of equipment operation, such as normal operation, overload operation, and failure occurrence, there are corresponding data records. The collected data also needs to be time synchronized and amplitude calibrated, and the processing method is the same as that for obtaining real-time data. For example, the sampling timestamp sequence of vibration waveform data, temperature gradient data, and current harmonic data is extracted respectively, the maximum common time interval is determined, the time axis is aligned, and then the preset sensor calibration coefficient set is called for amplitude compensation, and finally normalization is performed to generate a standardized historical monitoring data set.

[0104] Step S2102: extracting operation features from the standardized historical monitoring data set to generate a historical operation feature set.

[0105] In this embodiment, the operation feature extraction can be performed on the standardized historical monitoring data set, and the process is similar to the operation feature extraction of real-time data. For example, the time domain feature extraction is performed on the vibration waveform data to calculate the vibration effective value, peak factor and kurtosis coefficient; wavelet packet decomposition is performed to extract the preset frequency band energy proportion feature and frequency band entropy value feature. The temperature gradient data is processed by sliding window mean to generate temperature change rate feature and heat accumulation trend feature. The current harmonic data is processed by fast Fourier transform to extract the fundamental wave amplitude, harmonic distortion rate and characteristic subharmonic component amplitude ratio. The above features are combined to form a historical operation feature set.

[0106] Step S2103: marking a health status label at each time point based on the equipment maintenance record, wherein the health status label is quantified according to the ratio of the actual number of maintenance times to the standard maintenance cycle.

[0107] In this embodiment, the equipment maintenance record records the maintenance status of the equipment at different time points in detail. Based on the above records, the ratio of the actual number of maintenance times at each time point to the standard maintenance cycle is calculated to quantify the health status label. For example, the standard maintenance cycle stipulates that the equipment should be fully maintained every 1000 hours. At a certain time point, the equipment has been running for 5000 hours and has actually been repaired 3 times. The quantified value of the health status label at this time point is 3 / (5000 / 1000)=0.6. The value of the health status label is between 0 and 1, and the smaller the value, the better the health status of the equipment.

[0108] Step S2104: An abnormal risk level label is marked based on the fault event record, and the abnormal risk level label is graded according to the standard deviation of the characteristic fluctuation amplitude within N hours before the fault occurs.

[0109] In this embodiment, for the fault event record, various characteristic fluctuations of the equipment within N hours before the fault occurs are analyzed. For example, N=24 hours are selected to analyze the fluctuation range of characteristics such as vibration amplitude, temperature change, and current harmonics within these 24 hours. The standard deviation of the above characteristic fluctuation ranges is calculated, and the abnormal risk level is graded according to the size of the standard deviation. For example, when the standard deviation is less than a certain threshold, it is marked as a low risk level; when the standard deviation is within a certain range, it is marked as a medium risk level; when the standard deviation is greater than a higher threshold, it is marked as a high risk level.

[0110] Step S2105: labeling a performance degradation trend label based on the performance test data, wherein the performance degradation trend label is obtained by linear fitting according to the percentage of the output torque attenuation rate and the rated torque.

[0111] In this embodiment, the performance data of the device such as the output torque can be obtained by regularly performing performance tests on the device. The attenuation of the output torque over time is analyzed, and the percentage of the output torque attenuation rate to the rated torque is calculated. For example, the rated torque of the device is 1000N·m. Over a period of time, the output torque drops from the initial 1000N·m to 950N·m. After 500 hours, the output torque attenuation rate is (1000-950) / 500=0.1N·m / hour. The percentage of the rated torque is 0.1 / 1000×100%=0.01%. According to a series of such data points, linear fitting is performed to obtain a performance degradation trend label. For example, after multiple performance tests and data fitting, a linear equation is obtained, which can represent the relationship between the output torque attenuation rate and time, and the relevant parameters or slope of the linear equation and other information are used as the performance degradation trend label. Through the above steps, a multi-task training data set containing a historical operation feature set and its corresponding health status label, abnormal risk level label and performance degradation trend label is obtained.

[0112] Step S220: Initialize the convolutional neural network parameters of the shared feature extraction layer, the gated recurrent unit network parameters of the real-time health status prediction branch, the multi-head self-attention mechanism parameters of the abnormal risk classification branch, and the radial basis function network parameters of the performance degradation regression branch.

[0113] In this embodiment, for the convolutional neural network with a shared feature extraction layer, its parameters include the weight and bias of the convolution kernel. For example, assuming that the convolution kernel size is 3×3 and there are 16 convolution kernels, each convolution kernel has 3×3 weight values ​​and a bias value. The above parameters will be assigned random values ​​during initialization, but usually follow a certain distribution, such as a normal distribution or a uniform distribution. For example, the weight value will be randomly selected between -0.1 and 0.1, and the bias value will be initialized to 0.

[0114] The network parameters of the gated recurrent unit of the real-time health status prediction branch include the weights and biases of the update gate, reset gate, and output gate. For example, the size of the weight matrix Wz of the update gate is [input dimension, hidden dimension], the size of the weight matrix Wr of the reset gate is also [input dimension, hidden dimension], and the size of the weight matrix Wh of the output gate is [input dimension, hidden dimension]. Their weight values ​​are also randomly selected during initialization, and the bias values ​​are also initialized to appropriate small values.

[0115] The parameters of the multi-head self-attention mechanism for the abnormal risk classification branch include the projection matrix and the attention weights. For example, assuming there are 8 heads, the projection matrix of each head projects the input feature vector into a different subspace, and the weights of the projection matrix are randomly assigned at initialization. The attention weights are also assigned small random values ​​at initialization so that they can be adjusted according to the data during training.

[0116] The radial basis function network parameters of the performance degradation regression branch include the center vector and the width parameter. For each radial basis function, the dimension of the center vector is the same as the dimension of the input feature vector, and its value can be reasonably set according to the range of the input data during initialization, such as taking a value near the mean of the input data. The width parameter determines the width of the radial basis function and is also set to a suitable value, such as 0.5, during initialization. Through such initialization, the initial parameter setting is provided for the training of the multi-task state evaluation model.

[0117] Step S230: inputting the historical operation feature set into the initialized shared feature extraction layer to generate a historical shared implicit feature vector.

[0118] In this embodiment, the acquired historical operation feature set can be used as input and passed into the initialized shared feature extraction layer. The convolutional neural network of the shared feature extraction layer begins to process the historical operation feature set. For example, the historical operation feature set can be regarded as a multidimensional array containing multiple feature information such as vibration, temperature, and current. The convolution kernel of the convolutional neural network performs a sliding convolution operation on the multidimensional array, and each convolution calculates the dot product of the convolution kernel and the local feature, and adds a bias value. For example, for a 3×3 convolution kernel, a convolution calculation is performed with a 3×3 local feature area at a certain position to obtain a new feature value. After multiple convolution operations, the feature information is gradually extracted and compressed. Then a pooling operation, such as maximum pooling or average pooling, may be performed to further reduce the feature dimension and retain key features. For example, in maximum pooling, the maximum value is selected in a local area as the output after pooling. After a series of convolution and pooling operations, a historical shared implicit feature vector is finally generated, which contains the key information in the historical operation feature set.

[0119] Step S240: Input the historical shared implicit feature vector into each branch network respectively, calculate the mean square error between the output of the real-time health status prediction branch and the health status label, the cross entropy loss between the output of the abnormal risk classification branch and the abnormal risk level label, and the cosine similarity loss between the output of the performance degradation regression branch and the performance degradation trend label, and obtain the loss value of each branch.

[0120] In this embodiment, the generated historical shared implicit feature vectors may be input into the real-time health status prediction branch, the abnormal risk classification branch, and the performance degradation regression branch respectively.

[0121] In the real-time health status prediction branch, the gated recurrent unit network performs temporal dependency modeling based on the historical shared implicit feature vector and outputs a predicted real-time health status indicator. Assuming that the output prediction value is y_pred and the corresponding health status label is y_true, the mean square error is calculated by first calculating the square of the error between the predicted value and the true value of each sample, that is, (y_pred[i]-y_true[i])², where i represents the sample index. Then sum the squared errors of all samples and divide them by the number of samples to obtain the mean square error MSE=Σ(y_pred[i]-y_true[i])² / n, where n is the number of samples.

[0122] In the abnormal risk classification branch, the multi-head self-attention mechanism focuses on the key features of the historical shared implicit feature vector and outputs a probability distribution vector, which indicates the probability of the device being at different abnormal risk levels. Assume that the output probability distribution vector is P_pred, and the actual abnormal risk level label is a one-hot encoding vector P_true. The cross entropy loss is calculated by calculating -P_true[i]*log(P_pred[i]) for each category and then summing it over all categories, that is, the cross entropy loss CE=-Σ(P_true[i]*log(P_pred[i])).

[0123] In the performance degradation regression branch, the radial basis function network performs nonlinear mapping on the historical shared implicit feature vector and outputs a quantitative value of the performance degradation trend. Assuming the output value is z_pred and the performance degradation trend label is z_true, the cosine similarity loss is calculated by first calculating the dot product of the predicted value vector and the true value vector, dividing it by their modulus product, and then subtracting the cosine similarity value from 1. That is, cosine similarity Cosine_similarity=(z_pred·z_true) / (||z_pred||||z_true||), and cosine similarity loss CL=1-Cosine_similarity. Through the above calculations, the loss values ​​of each branch are obtained, which reflect the gap between the current prediction performance of each branch network and the true label.

[0124] Step S250: Dynamically adjust the parameter update amplitude of the shared feature extraction layer according to the back propagation gradient of the loss value of each branch, and give priority to optimizing the network parameters corresponding to the branch task with the highest loss value.

[0125] For example, step S250 may include:

[0126] Step S2501: In each training iteration, the loss values ​​of each branch are input into a dynamic weight allocator to generate a gradient weight coefficient for each branch.

[0127] In this embodiment, at each training iteration, the mean square error of the real-time health status prediction branch, the cross entropy loss of the abnormal risk classification branch, and the cosine similarity loss of the performance degradation regression branch are input into the dynamic weight allocator. The dynamic weight allocator calculates the gradient weighting coefficient of each branch according to the size of the above loss value. For example, a simple method can be used to calculate the proportion of each branch loss value to the total loss value as the gradient weighting coefficient. Assume that the mean square error of the real-time health status prediction branch is E1, the cross entropy loss of the abnormal risk classification branch is E2, and the cosine similarity loss of the performance degradation regression branch is E3, and the total loss value E=E1+E2+E3. Then the gradient weighting coefficient w1=E1 / E of the real-time health status prediction branch, the gradient weighting coefficient w2=E2 / E of the abnormal risk classification branch, and the gradient weighting coefficient w3=E3 / E of the performance degradation regression branch. The above gradient weighting coefficients reflect the relative importance of each branch loss in the total loss.

[0128] Step S2502: adjusting the gradient amplitude of the parameters of the shared feature extraction layer during the back propagation process according to the gradient weighting coefficient, so that the branch task with a higher loss value obtains a larger parameter update weight.

[0129] In this embodiment, during the back propagation process, the parameter update of the shared feature extraction layer depends on the gradient information of each branch. According to the calculated gradient weighting coefficient, the gradient amplitude of the shared feature extraction layer parameters is adjusted. For example, for the weight parameter of a convolution kernel of the shared feature extraction layer, its gradient was originally g. When adjusting, its gradient is multiplied by the corresponding gradient weighting coefficient. If the loss value of the abnormal risk classification branch is high and its gradient weighting coefficient w2 is large, then the gradient related to the abnormal risk classification branch will be amplified when it is back propagated to the shared feature extraction layer. In this way, branch tasks with higher loss values ​​can obtain larger parameter update weights in the shared feature extraction layer, thereby giving priority to optimizing the network parameters corresponding to the branch tasks and improving their prediction performance.

[0130] Step S2503: When the cross entropy loss of the abnormal risk classification branch does not decrease after N consecutive iterations, freeze the parameter update of the real-time health status prediction branch to enhance the feature extraction capability of the abnormal risk classification task.

[0131] In this embodiment, a threshold N is set, for example, N=5. When the cross entropy loss of the abnormal risk classification branch does not decrease in five consecutive iterations, it means that the current training process may not effectively optimize the abnormal risk classification task. At this time, in order to concentrate resources to optimize the abnormal risk classification task, the parameter update of the real-time health status prediction branch is frozen. That is, the gated recurrent unit network parameters of the real-time health status prediction branch are no longer adjusted according to the current error. In this way, the parameter update of the real-time health status prediction branch can be prevented from interfering with the optimization process of the abnormal risk classification task, so that the shared feature extraction layer can focus more on extracting features that are useful for abnormal risk classification, strengthen the feature extraction capability of the abnormal risk classification task, and improve its classification accuracy.

[0132] Step S2504: When the cosine similarity loss of the performance-degraded regression branch reaches a preset threshold, the parameter freezing state of all branches is released and the multi-task collaborative training mode is restored.

[0133] In this embodiment, a preset threshold is set, for example, when the cosine similarity loss of the performance degradation regression branch drops below 0.1. When the preset threshold is reached, it means that the performance degradation regression branch has achieved a good optimization effect. At this time, the parameter freezing state of all branches is released, including the previously frozen parameters of the real-time health status prediction branch and the abnormal risk classification branch. The multi-task collaborative training mode is restored so that each branch can update its parameters again according to its own error, and the shared feature extraction layer can perform comprehensive parameter adjustments based on the feedback information of all branches, thereby realizing collaborative optimization between multiple tasks and further improving the performance of the entire multi-task state evaluation model.

[0134] Step S260: Repeat the feature extraction and parameter updating process until the loss values ​​of all branches converge to a preset threshold range, thereby generating a trained multi-task state evaluation model.

[0135] In this embodiment, the process from step S230 to step S250 can be repeated continuously, that is, each time the historical operation feature set is input into the shared feature extraction layer to generate a historical shared implicit feature vector, and then the historical shared implicit feature vector is input into each branch network to calculate the loss value, and then the parameter update amplitude of the shared feature extraction layer is adjusted according to the loss value. In this process, the loss value of each branch is continuously monitored. A threshold interval is preset, for example, for mean square error, cross entropy loss and cosine similarity loss, the preset threshold interval is [0.05, 0.15]. When the mean square error of the real-time health status prediction branch, the cross entropy loss of the abnormal risk classification branch and the cosine similarity loss of the performance degradation regression branch all converge to the preset threshold interval, it means that the model has learned enough feature information, and the prediction performance of each branch has reached a relatively stable and satisfactory level. At this point, the training process ends, and a multi-task state evaluation model with training is generated. The multi-task state evaluation model can be used to accurately evaluate the state of the operation feature set of the target servo drive device, generate real-time health status indicators, abnormal risk levels and performance degradation trends, and provide a reliable basis for subsequent equipment maintenance optimization strategies.

[0136] Figure 2 A schematic diagram of exemplary hardware and software components of a servo drive system state monitoring system 100 based on multi-sensor fusion that can implement the concept of the present application is shown in some embodiments of the present application. For example, the processor 120 can be used in the servo drive system state monitoring system 100 based on multi-sensor fusion and used to perform the functions in the present application.

[0137] The servo drive system condition monitoring system 100 based on multi-sensor fusion can be a general-purpose server or a special-purpose server, both of which can be used to implement the servo drive system condition monitoring method based on multi-sensor fusion of the present application. Although only one server is shown in the present application, for convenience, the functions described in the present application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0138] For example, the servo drive system condition monitoring system 100 based on multi-sensor fusion may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and storage media 140 in different forms, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the servo drive system condition monitoring system 100 based on multi-sensor fusion may also include program instructions stored in ROM, RAM, or other types of non-temporary storage media, or any combination thereof. The method of the present application can be implemented according to the above-mentioned program instructions. The servo drive system condition monitoring system 100 based on multi-sensor fusion also includes an input / output (I / O) interface 150 between the computer and other input / output devices.

[0139] For ease of explanation, only one processor is described in the servo drive system condition monitoring system 100 based on multi-sensor fusion. However, it should be noted that the servo drive system condition monitoring system 100 based on multi-sensor fusion in the present application may also include multiple processors, so the steps performed by one processor described in the present application may also be performed jointly or individually by multiple processors. For example, if the processor of the servo drive system condition monitoring system 100 based on multi-sensor fusion executes step A and step B, it should be understood that step A and step B may also be executed jointly by two different processors or individually in one processor. For example, the first processor executes step A, the second processor executes step B, or the first processor and the second processor execute steps A and B together.

[0140] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer executable instructions are preset. When a processor executes the computer executable instructions, the servo drive system state monitoring method based on multi-sensor fusion as described above is implemented.

[0141] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, various features are sometimes combined into one embodiment, drawing or description thereof.

Claims

1. A servo drive system state monitoring method based on multi-sensor fusion, characterized in that: The method comprises: Acquire a multi-source sensor data set of a target servo drive device, wherein the multi-source sensor data set includes vibration waveform data, temperature gradient data, and current harmonic data; Performing time synchronization and amplitude calibration processing on the multi-source sensor data set to generate a standardized monitoring data set; Calling a dynamic coding model to extract operation features from the standardized monitoring data set to generate an operation feature set of a target device; Performing a status assessment on the operation feature set based on a multi-task status assessment model to generate a real-time health status indicator, an abnormal risk level, and a performance degradation trend of the target device; An equipment maintenance optimization strategy is generated according to the real-time health status indicators, abnormal risk levels and performance degradation trends, and the equipment maintenance optimization strategy is fed back to the operation and maintenance control platform to trigger active maintenance operations.

2. The servo drive system state monitoring method based on multi-sensor fusion according to claim 1 is characterized in that: The performing time synchronization and amplitude calibration processing on the multi-source sensor data set to generate a standardized monitoring data set includes: Extracting a first sampling timestamp sequence of the vibration waveform data, a second sampling timestamp sequence of the temperature gradient data, and a third sampling timestamp sequence of the current harmonic data; Determine the maximum common time interval between the first sampling timestamp sequence, the second sampling timestamp sequence and the third sampling timestamp sequence, perform time axis alignment processing on the vibration waveform data, the temperature gradient data and the current harmonic data based on a linear interpolation algorithm, and generate a time synchronization data set; Calling a preset sensor calibration coefficient set to perform amplitude compensation processing on the vibration waveform data, temperature gradient data and current harmonic data in the time synchronization data set, wherein the sensor calibration coefficient set includes a frequency response compensation factor of the vibration sensor, a nonlinear error correction factor of the temperature sensor and a phase offset correction parameter of the current sensor; The amplitude-compensated vibration waveform data, temperature gradient data and current harmonic data are normalized to generate a standardized monitoring data set.

3. The servo drive system state monitoring method based on multi-sensor fusion according to claim 1 is characterized in that: The calling of the dynamic coding model to extract the operation characteristics of the standardized monitoring data set to generate the operation characteristic set of the target device includes: Extracting time-domain features of the vibration waveform data in the standardized monitoring data set to generate a vibration effective value, a peak factor, and a kurtosis coefficient; Performing wavelet packet decomposition processing on the vibration waveform data to extract the energy proportion characteristics and frequency band entropy value characteristics of the preset frequency band; Performing sliding window mean processing on the temperature gradient data to generate temperature change rate characteristics and heat accumulation trend characteristics; Performing fast Fourier transform processing on the current harmonic data to extract fundamental wave amplitude, harmonic distortion rate and characteristic subharmonic component amplitude ratio; The vibration effective value, peak factor, kurtosis coefficient, preset frequency band energy proportion characteristics, frequency band entropy value characteristics, temperature change rate characteristics, heat accumulation trend characteristics, fundamental wave amplitude, harmonic distortion rate and characteristic subharmonic component amplitude ratio are input into the dynamic coding model, and dynamic association coding is performed through a multi-layer cross-attention mechanism to generate an operation feature set.

4. The servo drive system state monitoring method based on multi-sensor fusion according to claim 1 is characterized in that: The state evaluation of the operation feature set based on the multi-task state evaluation model to generate the real-time health status indicator, abnormal risk level and performance degradation trend of the target device includes: Inputting the running feature set into a shared feature extraction layer to generate a shared latent feature vector; Inputting the shared implicit feature vector into the real-time health status prediction branch, the abnormal risk classification branch and the performance degradation regression branch respectively; In the real-time health status prediction branch, the shared implicit feature vector is modeled for time series dependency based on a gated recurrent unit network, and a real-time health status indicator is output; In the abnormal risk classification branch, the shared implicit feature vector is focused on key features based on a multi-head self-attention mechanism, and a probability distribution of the abnormal risk level is output; In the performance degradation regression branch, nonlinear mapping is performed on the shared implicit feature vector based on a radial basis function network to output a quantized value of the performance degradation trend; The weight allocation ratio of the shared feature extraction layer is dynamically adjusted according to the output results of the real-time health status prediction branch, the abnormal risk classification branch and the performance degradation regression branch to achieve multi-task collaborative optimization.

5. The servo drive system state monitoring method based on multi-sensor fusion according to claim 1 is characterized in that: Generating an equipment maintenance optimization strategy according to the real-time health status indicator, abnormal risk level and performance degradation trend includes: Establish health status indicator threshold intervals, abnormal risk level threshold intervals, and performance degradation trend slope thresholds; When the real-time health status indicator is lower than the lower limit of the health status indicator threshold interval and the abnormal risk level exceeds the first risk threshold, an immediate shutdown maintenance instruction is generated; When the performance degradation trend slope exceeds a slope threshold and the abnormal risk level is within a second risk threshold interval, generating a preventive maintenance plan and a spare parts replacement suggestion; When the real-time health status indicator is in the middle of the health status indicator threshold interval and the abnormal risk level is lower than the third risk threshold, generating an operation parameter optimization adjustment plan; The immediate shutdown and maintenance instructions, preventive maintenance plans, spare parts replacement suggestions and operating parameter optimization adjustment plans are integrated into an equipment maintenance optimization strategy.

6. The servo drive system state monitoring method based on multi-sensor fusion according to claim 3 is characterized in that: The wavelet packet decomposition process is performed on the vibration waveform data to extract the preset frequency band energy proportion characteristics and frequency band entropy value characteristics, including: The Daubechies wavelet basis function is selected to perform 5-layer wavelet packet decomposition on the vibration waveform data to obtain 32 frequency band node coefficients; Calculate the energy value of each frequency band node, and divide the energy value of each frequency band by the total energy to obtain the preset frequency band energy ratio feature; Calculate the permutation entropy of the coefficient sequence of each frequency band node to obtain the frequency band entropy value feature; The characteristic frequency bands whose energy proportion exceeds the set threshold are screened, and their corresponding frequency band entropy value features and energy proportion features are added to the operation feature set.

7. The servo drive system state monitoring method based on multi-sensor fusion according to claim 4 is characterized in that: The method dynamically adjusts the weight distribution ratio of the shared feature extraction layer according to the output results of the real-time health status prediction branch, the abnormal risk classification branch, and the performance degradation regression branch to achieve multi-task collaborative optimization, including: Calculate the prediction error rate of the real-time health status prediction branch, the cross entropy loss value of the abnormal risk classification branch, and the mean square error of the performance degradation regression branch; The prediction error rate, cross entropy loss value and mean square error are normalized to obtain a weight adjustment factor for each branch; Dynamically update the channel attention weights of each convolution kernel in the shared feature extraction layer according to the weight adjustment factor; When the weight adjustment factor of the abnormal risk classification branch exceeds the preset alarm threshold, the parameter update of the performance degradation regression branch is frozen, and the abnormal risk classification task is optimized first.

8. The servo drive system state monitoring method based on multi-sensor fusion according to claim 5 is characterized in that: The generation of preventive maintenance plans and spare parts replacement recommendations includes: Calculate the remaining service life prediction value based on the performance degradation trend slope; Query the equipment maintenance knowledge graph to match the historical maintenance case set corresponding to the remaining service life prediction value; Extract optimal maintenance time windows, spare parts model replacement records, and labor cost data from historical maintenance cases; Based on the fuzzy decision-making algorithm, multi-objective optimization of maintenance time windows, spare parts inventory status and production plans is carried out to generate the time nodes of the preventive maintenance plan, the required spare parts list and the human resource allocation plan.

9. The servo drive system state monitoring method based on multi-sensor fusion according to claim 8 is characterized in that: The multi-objective optimization of the maintenance time window, spare parts inventory status and production plan based on the fuzzy decision algorithm is performed to generate the time nodes of the preventive maintenance plan, the required spare parts list and the human resource allocation plan, including: Converting the maintenance time window into a time overlap conflict matrix, mapping the spare parts inventory status into a spare parts availability vector, and parsing the production plan into a maintenance operation time constraint sequence; Based on the time overlap conflict matrix, spare parts availability vector and maintenance operation time constraint sequence, a multi-objective optimization space is constructed, and a set of candidate maintenance solutions is generated using a Pareto frontier search algorithm; Performing time conflict resolution processing on the candidate maintenance solution set, eliminating candidate solutions that conflict with the maintenance operation time constraint sequence, and generating a conflict-free candidate maintenance solution set; According to the spare parts availability vector, the set of conflict-free candidate maintenance solutions is sorted by spare parts matching degree, and the top k candidate solutions with the highest spare parts matching degree are selected; The first k candidate solutions are input into the fuzzy decision maker, and the time nodes, required spare parts list and human resource allocation plan of the preventive maintenance plan are generated according to the preset maintenance priority rules.

10. A servo drive system state monitoring system based on multi-sensor fusion, characterized in that: It includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the servo drive system state monitoring method based on multi-sensor fusion as described in any one of claims 1 to 9.

Images