An adaptive processing method, device, and medium for multi-parameter monitoring signals of an electric motor drive system.

By adopting an adaptive processing method, the problems of signal processing adaptability and anomaly handling in motor drive systems are solved, achieving efficient signal processing and fault feature extraction under complex working conditions, and improving fault early warning capability and signal timing consistency.

CN123068779APending Publication Date: 2026-11-10SOUTHWEST CHINA RES INST OF ELECTRONICS EQUIP
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611105814.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-24
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing motor drive system condition monitoring suffers from problems such as weak signal processing adaptability, homogenized anomaly handling, insufficient ability to extract weak features, poor timing consistency of multi-source signals, and lack of dynamic optimization mechanism.

Method used

An adaptive processing method for multi-parameter monitoring signals of a motor drive system is adopted, which includes synchronous acquisition of system signals, hierarchical determination of interference level, adaptive wavelet threshold denoising, quantitative classification to identify anomaly types, accurate anomaly removal, Kalman filtering with residual discrimination for smoothing, and signal synchronous calibration through zero-crossing feature points to form a closed-loop iterative optimization.

Benefits of technology

It achieves efficient and differentiated processing of multi-parameter signals under complex operating conditions, accurately distinguishes abnormal types, retains weak fault characteristics, improves early fault warning capability and signal timing consistency, and has the ability to self-identify operating conditions and self-adapt to interference.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123068779A_ABST
    Figure CN123068779A_ABST
Patent Text Reader

Abstract

This invention belongs to the field of motor drive equipment condition monitoring and signal processing technology, specifically involving an adaptive processing method, device, and medium for multi-parameter monitoring signals of motor drive systems. Addressing the technical problems in existing technologies such as weak adaptability to operating conditions, homogenized anomaly handling, insufficient weak feature extraction capabilities, poor timing consistency of multi-source signals, and lack of dynamic optimization mechanisms, this invention proposes a signal processing method based on a heterogeneous signal independent quantization and hierarchical denoising, automatic identification and differentiated processing of multiple anomalies, synchronous calibration of multi-parameter timing, and a closed-loop iterative optimization architecture. The algorithm of this invention can be modularly embedded and deployed, supporting long-term online monitoring and fault diagnosis of motor drive systems.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of motor drive equipment condition monitoring and signal processing technology, specifically involving an adaptive processing method, equipment and medium for multi-parameter monitoring signals of a motor drive system. Background Technology

[0002] Currently, as the core power control and transmission component of equipment, the condition monitoring of motor drive systems mostly adopts a simple data acquisition scheme using a single sensor or multiple sensors, which has the following limitations in the signal processing stage: 1) Traditional solutions often use single algorithms such as fixed threshold wavelet denoising and conventional Kalman filtering, which are difficult to adapt to the complex working conditions of the transmission system in the field, dynamic interference changes of variable speed and load, and are prone to problems of insufficient filter strength matching.

[0003] 2) Conventional signal processing methods are too general and cannot specifically distinguish between various abnormal signal types such as random outliers, electromagnetic periodic interference, and temperature baseline drift, making it difficult to adapt to the differentiated processing requirements of different interferences.

[0004] 3) Current high-frequency response signals, vibration and shock signals, and temperature slow-change drift signals are heterogeneous signals. Existing technologies lack differentiated identification and precise processing strategies, and have limited ability to extract weak fault features.

[0005] 4) The independent timing of multi-source heterogeneous signal acquisition is prone to time sequence deviation, which affects the accuracy of back-end feature correlation analysis and state judgment.

[0006] 5) Traditional signal processing is mostly an open-loop, single-operation mode, which cannot dynamically iterate and optimize according to the field conditions and interference environment, and its adaptability and stability are limited under long-term operation. Summary of the Invention

[0007] This invention addresses the limitations of existing multi-parameter monitoring signal processing techniques for motor drive systems, including weak adaptability to operating conditions, homogenized anomaly handling, insufficient weak feature extraction capabilities, poor timing consistency of multi-source signals, and lack of dynamic optimization mechanisms. In its first aspect, it provides an adaptive processing method for multi-parameter monitoring signals in motor drive systems, comprising: Step 1: Synchronously acquire the first signals of the three channels of the system, and calculate the interference factor of the first signal of each channel. IF Based on the configured interference determination threshold, according to the interference factor IF The interference level of the first signal in each channel is determined as strong interference, moderate interference, or weak interference. Step 2: Perform improved adaptive wavelet threshold denoising on the first signal with strong interference level to obtain the second signal; for the first signal with moderate interference level, skip the improved adaptive wavelet threshold denoising and use the original first signal as the second signal; for the first signal with weak interference level, skip steps 3 and 4 and use the original first signal as the third signal. Step 3: Calculate the normal threshold range based on the second signal of each channel, and identify the abnormal type of each second signal based on the pre-defined quantization classification and identification mechanism. The abnormal types include random outliers, periodic electromagnetic interference and temperature drift. Step 4: Based on the second signal of each anomaly type, perform anomaly removal on the second signal of each anomaly type according to the pre-defined quantification anomaly removal mechanism to obtain the third signal; Step 5: Perform Kalman filtering with residual discrimination on each third signal to smooth the signal and output a smoothed signal. During the filtering process, the filtering parameters are adaptively updated based on the comparison between the filtering residual and the preset threshold. Step 6: Perform synchronous calibration on each smoothed signal to obtain the fourth signal; Step 7: Feed back the filter residuals and filter parameters from Step 5 and the periodic interference main frequency extracted in Step 4 to Step 1 to dynamically update the interference judgment threshold for the next cycle, forming a closed-loop iterative optimization.

[0008] Further, in step 1, the interference factor of the first signal in each channel is calculated respectively. IF According to the interference factor IF The interference level of the first signal in each channel is determined as strong interference, moderate interference, or weak interference, including: In one processing cycle, the sensor acquires the sampled values ​​of each channel of the system based on a preset sampling frequency as the first signal, and calculates the interference factor based on the sampled values ​​of each channel, the signal mean, and the number of sampling points. IF ,include:

[0009] In the formula, x ( n ) is the first single-channel signal n A momentary sample value, The average value of the signal amplitude. N This represents the number of signal sampling points. Preset strong interference threshold IF H and low interference threshold IF L And the interference factor calculated for each channel IF With respect to the preset strong interference threshold IFH and low interference threshold IF L By comparison, the interference levels of each channel signal are determined to be strong interference, moderate interference, and weak interference based on the comparison results; if IF ≥ IF H The interference level of the first signal on the corresponding channel is determined to be strong interference. if IF L ≤ IF < IF H The interference level of the first signal on the corresponding channel is determined to be moderate interference. if IF < IF L The interference level of the first signal on the corresponding channel is determined to be weak interference.

[0010] Further, in step 2, the first signal with a strong interference level is subjected to improved adaptive wavelet threshold denoising to obtain the second signal, including: The first signal, which has a strong interference level, is decomposed into low-frequency approximation coefficients using wavelet transform. and high-frequency detail coefficients of each layer Constructing an adaptive wavelet threshold The adaptive wavelet threshold The calculation formula includes:

[0011] in σ For noise variance, j The wavelet decomposition level; Based on an adaptive wavelet threshold, a hierarchical processing rule for wavelet coefficients is established, which includes high-frequency detail coefficients at each level. Output the correction coefficients after threshold processing based on the hierarchical processing rules. The correction coefficients and the low-frequency approximation coefficients are combined. Perform inverse wavelet transform reconstruction to obtain the second signal.

[0012] Furthermore, the step of formulating wavelet coefficient hierarchical processing rules based on adaptive wavelet thresholds, and processing high-frequency detail coefficients at each level... Output the correction coefficients after threshold processing based on the hierarchical processing rules. ,include: If the absolute value of the high-frequency detail factor satisfies Then the high-frequency detail coefficients will be... Set it directly to zero as the correction factor ; If the absolute value of the high-frequency detail factor satisfies Then the high-frequency detail coefficients will be... The correction factor is obtained by multiplying by a scaling factor and then compressing the product. ; If the absolute value of the high-frequency detail factor satisfies If a soft threshold shrinkage process is applied as a correction coefficient, then the soft threshold shrinkage process is performed through... Calculate the correction factor , where sgn( The parentheses are symbolic functions. If the value of the variable inside the parentheses is greater than 0, the function value is 1; if the value of the variable inside the parentheses is equal to 0, the function value is 0; if the value of the variable inside the parentheses is less than 0, the function value is -1.

[0013] The beneficial effects of the above steps are as follows: Unlike the traditional single hard / soft threshold, the three-stage hierarchical constraint can directly filter out small high-frequency background noise below half the threshold, compress and retain the critical weak components between half and full thresholds, and apply soft threshold shrinkage processing to the effective signal features above the threshold. While effectively filtering out invalid noise, it retains the sensitive features of impact faults in motor drive equipment to the maximum extent, taking into account both noise reduction and fault feature retention.

[0014] Further, in step 3, the normal threshold range of the time-domain amplitude of the second signal for each channel is calculated, and the abnormal type of each second signal is identified based on a pre-defined quantization classification and identification mechanism, including: The average value of the second signal amplitude on each channel is calculated as follows: μ and standard deviation σ And based on the mean and standard deviation, the normal threshold range of the second signal is calculated as [ , ],in , ; A pre-defined quantitative classification and identification mechanism is established, and the second signal is output as an anomaly type of random outlier, periodic electromagnetic interference, temperature drift and normal signal based on the pre-defined quantitative classification and identification mechanism. The pre-defined quantitative classification and identification mechanisms include an identification mechanism based on the duration of instantaneous anomalies, an identification mechanism based on periodic repetition, and an identification mechanism based on slowly varying drift characteristics, including: Acquire all sampling points on any channel within the same processing cycle and arrange them in chronological order of acquisition time. Record all sampling points whose signal amplitude exceeds the normal threshold range after arrangement as outliers, and count the number of outliers. ; The mechanism for identifying the duration of the instantaneous anomaly includes: if =1, and the second signal before and after the collection time of the anomaly point is within the normal threshold range, then the anomaly type of the signal is determined to be a random outlier. The identification mechanism based on periodic repetition includes: if Within the set L signal observation windows, all anomalies are counted. Anomalies with adjacent acquisition times are grouped together to form an anomaly signal segment. All anomaly signal segments are arranged in order of acquisition time. If the percentage of the relative deviation between the actual time interval between the characteristic extreme points of two adjacent anomaly data segments and the preset fixed theoretical period T is less than a preset percentage threshold, then the anomaly type of the signal is determined to be periodic electromagnetic interference. Here, the characteristic extreme point is a local maximum or local minimum value in the anomaly data segment, and the fixed theoretical period T is the interference main frequency extracted from the signal frequency domain. The reciprocal of; The identification mechanism based on slowly varying drift features includes: if If the value is 0, then starting from the second sampling point, calculate the difference between the temperature value of the current sampling point and the temperature value of the second signal at the previous moment as the first-order difference of the current sampling point. If a preset number of consecutive... Nc The first-order differences of each sampling point all retain the same sign and are continuous. Nc The cumulative temperature change at each sampling point is greater than a preset temperature threshold. T diff If the signal is determined to be an anomaly of temperature drift, then the cumulative temperature change is... Nc The absolute value of the temperature difference between the first and last sampling points in a sampling point; Furthermore, the cumulative temperature change includes: across all preset quantities Nc The absolute value of the temperature difference between the first and last sampling points in a sampling point group.

[0015] Furthermore, second signals that do not conform to any identification mechanism are judged as normal signals, and the original normal signals are output as the third signals.

[0016] Further, in step 4, the third signal is obtained by removing anomalies from the second signal of each anomaly type based on a pre-defined quantization anomaly removal mechanism, including: A quantitative anomaly removal mechanism is pre-defined, and the second signals of each anomaly type are removed based on the pre-defined quantitative anomaly removal mechanism to obtain the third signal; The pre-defined quantitative anomaly removal mechanism includes: a precise removal mechanism based on random outliers, a precise removal mechanism based on periodic electromagnetic interference, and a precise removal mechanism based on temperature drift. A precise outlier removal mechanism based on random outliers: Centering on the index k of an outlier of the random outlier type, several sampling points before and after the outlier's sampling time are simultaneously acquired as initial candidate points. Outliers present in all initial candidate points are removed, and a predetermined number of remaining candidate points are retained as valid candidate points, forming a valid candidate point set Ω. The weight of each valid candidate point in set Ω is then calculated. And through the weighted average formula Calculate the corrected replacement value and replace the signal amplitude at the original anomaly point with the corrected replacement value, wherein... Let be the signal amplitude of a valid candidate point, and i be the index of the valid candidate point. The time series distance between the outlier index and the candidate index; A precise elimination mechanism based on periodic electromagnetic interference: All anomalous signal segments of periodic interference type are acquired. Frequency domain components are analyzed using Fourier spectrum analysis. After eliminating the power frequency fundamental component, narrowband interference frequencies with amplitudes greater than three times the mean of the remaining spectrum are selected from the remaining spectrum. A second-order bandstop notch filter is designed based on the narrowband interference master frequency, with a stopband range of [missing information]. The abnormal signal segment of periodic electromagnetic interference is processed by narrowband filtering based on the second-order band-stop notch wave, and the signal after filtering out periodic interference is output as the third signal. Furthermore, when there are multiple interfering frequencies in the second signal, multiple band-stop notch filters are cascaded to filter out the interfering frequencies of each cycle in turn.

[0017] A precise elimination mechanism based on temperature drift is as follows: A continuous sampling point interval identified as a temperature drift type is defined as a drift interval; the temperature values ​​of all sampling points within the drift interval are extracted to form a temperature sequence to be fitted; a quadratic polynomial least squares fitting is performed on the temperature sequence to be fitted to obtain a slow drift trend term; the slow drift trend term is subtracted from the temperature values ​​in the temperature sequence to be fitted to obtain a calibrated temperature signal, and the calibrated temperature signal is used as the third signal.

[0018] Further, in step 5, the Kalman filtering smoothing process with residual discrimination is performed on each third signal to output a smoothed signal. During the filtering process, the filtering parameters are adaptively updated based on the comparison result between the filtering residual and the preset threshold, including: Using a standard discrete linear Kalman filter recursive framework, the third signal is input into the recursive framework, and the observed signal at the current time k is... Once acquired, and before updating the observations at the current moment, the optimal estimate based on the previous moment is extracted from the standard recursive framework. The generated one-step state prediction value ; The observed signal and the predicted state value of the first step Difference to generate filter residuals ,in, H The state matrix of the Kalman filter framework; when If the signal smoothing is successful, the optimal state estimate obtained after updating the observation at the current time is directly output. As the final smoothed signal; when When it is determined that there is residual strong interference, the Kalman filter model process noise covariance Q and observation noise covariance R are adaptively updated, and the filtering iteration is re-executed on the third signal to recalculate the filtering residual. Set a maximum number of iterations When the detected filter residual is greater than Enter adaptive update mode; increment the iteration counter by 1 after each iteration. If the number of iterations reaches Afterwards, the calculated filter residual was still greater than If the system determines that the current signal belongs to an extreme strong transient interference or a sensor hardware failure, it will abandon further iteration and force the output of the latest optimal estimate. X ( k | k As a smoothing result, a filter anomaly alarm flag is generated and saved simultaneously for the smoothed signal output at this time. The adaptive update of the Kalman filter model process includes the noise covariance Q and the observation noise covariance R, which include: Calculate the residual deviation factor β , ; The updated process noise covariance Q is: Q_{new} = β×Q_{current}, where Q_{new} is the updated process noise covariance and Q_{current} is the current process noise covariance; Using the updated Q_{new} and the current observation noise covariance R Re-execute the Kalman filter iteration and recalculate the Kalman gain. K ( k And update the optimal state estimate. X ( k | k Then, calculate the filter residual again. If the filter residual after the re-iteration is still greater than Then, the observation noise covariance R is also appropriately amplified to R_{new} = 1.2×R, and the Kalman filter iteration steps are re-executed with the updated R_{new} until the filter residual meets the convergence condition or reaches the maximum number of iterations.

[0019] Further, in step 6, the smoothing signal undergoes synchronous calibration to obtain a fourth signal, including: Zero-crossing feature points of the smoothed signal on the reference channel are extracted as reference timestamps. The actual timestamps of the smoothed signals on other channels within the same processing cycle are read, and the time difference Δ between the actual timestamps of the smoothed signals on other channels and the reference timestamps is calculated. t ; If Δ t If the value is greater than a preset threshold, the signal amplitude and actual timestamp of the smoothed signal adjacent to the acquisition time of the smoothed signal on the other channels are obtained. The theoretical amplitude of the signal at the reference timestamp is calculated and reconstructed using a linear interpolation formula. The theoretical amplitude is then used to replace the original sampled value of the smoothed signal on the other channels at the actual timestamp to achieve precise alignment. If Δ t If the value is not greater than a preset threshold, the actual timestamp of the smoothed signal on the other channels is directly and forcibly rewritten to the reference timestamp to complete the physical alignment of the time dimension. Furthermore, the reference channel is a current signal channel, the zero-crossing feature point is the zero-crossing feature point of the current signal, and the other channels include a vibration signal channel and a temperature signal channel.

[0020] Furthermore, the parameters for the signal processing after calibration and synchronization include: noise standard deviation. σ Interference detection threshold, periodic interference main frequency The filter coefficients and other parameters are stored in memory. When adaptively processing the signal in the next cycle, the updated noise standard deviation is used. σ The interference level determination range and process parameters for the next cycle are dynamically corrected to form a continuous iterative optimization closed loop, thereby improving the stability and accuracy of signal processing under complex working conditions in the long term.

[0021] Furthermore, the use of the updated noise standard deviation σ Dynamically adjust the interference level determination range and process parameters for the next cycle to form a continuous iterative optimization closed loop, including: Calculate the current periodic noise standard deviation σ new The standard deviation of the noise from the preset reference noise level σ base The deviation rate is used to dynamically amplify or reduce the strong interference threshold I of the next cycle according to a preset adjustment coefficient. FH and low interference threshold I FL This forms a closed-loop iterative optimization.

[0022] In a second aspect, the present invention also provides an electronic device, comprising: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the above-described adaptive processing method for multi-parameter monitoring signals of the motor drive system.

[0023] In a third aspect, the present invention also provides a computer-readable storage medium including a computer program, which, when executed on a computer, causes the computer to perform the above-described adaptive processing method for multi-parameter monitoring signals of a motor drive system.

[0024] The beneficial effects of this invention are: (i) Interference classification and adaptive precision processing: The interference of the three signals is independently quantified and classified, and different noise reduction strategies are matched for strong, medium and weak interference intensities. High-frequency random noise and electromagnetic harmonic interference are thoroughly suppressed, perfectly adapting to complex field conditions such as heavy load, alternating speed and strong electromagnetic interference of motor drive system.

[0025] (II) Refined differentiation and processing of multiple types of anomalies: Through quantitative criteria such as the number of consecutive out-of-limit points, periodic interval error, and continuity of differential trend, the three typical anomalies of single-point outliers, periodic electromagnetic interference, and temperature baseline drift are accurately distinguished. Dedicated processing algorithms are provided to avoid feature loss caused by homogenized processing.

[0026] (III) High-fidelity extraction of weak fault features: Through optimization design such as three-segment wavelet threshold constraint, inverse distance weighted neighborhood correction, and baseline trend term separation and deduction, interference is efficiently filtered out while the early weak impact and fault feature components of bearings, gears and motors are fully preserved, thereby improving the early warning capability of faults.

[0027] (iv) High-precision timing unification of multi-source signals: Relying on the high-speed response zero-crossing feature point as the global time anchor point, the heterogeneous signals are precisely aligned at the sub-millisecond level, eliminating the data analysis error caused by timing offset and ensuring the accuracy of multi-source feature fusion and judgment.

[0028] (V) Closed-loop iterative self-adaptive optimization: The first closed-loop feedback mechanism for signal processing is used to dynamically update the judgment benchmark and filtering parameters using real-time effective data from the field. It has the ability to identify working conditions and adapt to interference, and its robustness and stability during long-term continuous operation are significantly improved.

[0029] (vi) Strong engineering feasibility: The overall algorithm is modular and lightweight with low computational load. It can be directly embedded and deployed in the local monitoring controller without additional hardware modification, and is suitable for large-scale online monitoring application scenarios of motor drive equipment. Attached Figure Description

[0030] Figure 1 A flowchart of an adaptive processing method for multi-parameter monitoring signals in a motor drive system provided by the present invention; Figure 2 This is a flowchart of an adaptive processing method for multi-parameter monitoring signals of a motor drive system provided in this embodiment. Detailed Implementation

[0031] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0032] Figure 1 The diagram shown is a flowchart of an adaptive processing method for multi-parameter monitoring signals in a motor drive system provided by the present invention. Figure 2 The diagram shown is a flowchart of an adaptive processing method for multi-parameter monitoring signals in a motor drive system provided in this embodiment. Please refer to... Figure 1 , Figure 2 As shown, the present invention provides an adaptive processing method for multi-parameter monitoring signals of a motor drive system, comprising: Step 1: Synchronously acquire the first signals of the three channels of the system, and calculate the interference factor of the first signal of each channel. IF Based on the configured interference determination threshold, according to the interference factor IF The interference level of the first signal in each channel is determined as strong interference, moderate interference, or weak interference. Step 2: Perform improved adaptive wavelet threshold denoising on the first signal with strong interference level to obtain the second signal; for the first signal with moderate interference level, skip the improved adaptive wavelet threshold denoising and use the original first signal as the second signal; for the first signal with weak interference level, skip steps 3 and 4 and use the original first signal as the third signal. Step 3: Calculate the normal threshold range based on the second signal of each channel, and identify the abnormal type of each second signal based on the pre-defined quantization classification and identification mechanism. The abnormal types include random outliers, periodic electromagnetic interference and temperature drift. Step 4: Based on the second signal of each anomaly type, perform anomaly removal on the second signal of each anomaly type according to the pre-defined quantification anomaly removal mechanism to obtain the third signal; Step 5: Perform Kalman filtering with residual discrimination on each third signal to smooth the signal and output a smoothed signal. During the filtering process, the filtering parameters are adaptively updated based on the comparison between the filtering residual and the preset threshold. Step 6: Perform synchronous calibration on each smoothed signal to obtain the fourth signal; Step 7: Feed back the filter residuals and filter parameters from Step 5 and the periodic interference main frequency extracted in Step 4 to Step 1 to dynamically update the interference judgment threshold for the next cycle, forming a closed-loop iterative optimization.

[0033] In this embodiment, a monitoring and testing platform for the motor drive system is built, and temperature sensors, vibration sensors, and current sensors are set up to collect the raw signals of the temperature channel, vibration channel, and current channel of the system respectively using an embedded acquisition module. The sampling frequency is 10Hz for temperature and 1kHz for current / vibration. Four operating conditions of the motor drive system are simulated: no-load, variable load, alternating speed, and strong electromagnetic interference. Various real-world scenarios such as slight bearing wear, motor electromagnetic interference, and slow rise and fall of ambient temperature are also simulated.

[0034] Further, in step 1, the interference factor of the first signal in each channel is calculated respectively. IF Based on the interference factor IF, the interference level of the first signal in each channel is determined to be strong interference, moderate interference, or weak interference, including: In one processing cycle, the sensor acquires the sampled values ​​of each channel of the system based on a preset sampling frequency as the first signal, and calculates the interference factor based on the sampled values ​​of each channel, the signal mean, and the number of sampling points. IF ,include:

[0035] In the formula, x ( n ) is the first single-channel signal n A momentary sample value, The average value of the signal amplitude. N This represents the number of signal sampling points. Preset strong interference threshold IF H and low interference threshold IF L The interference factor IF calculated for each channel is then compared with the preset strong interference threshold. IF H and low interference threshold IF L By comparison, the interference levels of each channel signal are determined to be strong interference, moderate interference, and weak interference based on the comparison results; if IF ≥ IF HThe interference level of the first signal on the corresponding channel is determined to be strong interference. if IF L ≤ IF < IF H The interference level of the first signal on the corresponding channel is determined to be moderate interference. if IF < IF L The interference level of the first signal on the corresponding channel is determined to be weak interference.

[0036] In this embodiment IF H The preset initial value is 0.35. IF L The initial value is 0.1. In this embodiment, the electromagnetic interference of the current signal is relatively strong, and the single-channel signal interference is due to... IF =0.42, which is considered strong interference.

[0037] Further, in step 2, the first signal with a strong interference level is subjected to improved adaptive wavelet threshold denoising to obtain the second signal, including: In this embodiment, considering the nonlinear and impulsive characteristics of vibration and current mixed signals in the motor drive system, the present invention preferably uses the db4 wavelet basis to perform multi-level wavelet decomposition, decomposing the original signal into low-frequency approximation coefficients. High-frequency detail coefficients of each layer (Those skilled in the art can replace it with similar tightly supported wavelet bases such as sym8 or coif5 according to actual working conditions, all of which fall within the protection scope of this invention). The larger the wavelet decomposition level j, the higher the corresponding frequency. The lower the value, the better. Electromagnetic interference and random pulse noise in motor drive systems are concentrated in the high-frequency range; therefore, the low-frequency approximation coefficient is... Fully preserved to maintain signal baseline and trend, only high-frequency detail coefficients at each layer are preserved. Perform threshold shrinkage and noise reduction processing; The first signal, which has a strong interference level, is decomposed into low-frequency approximation coefficients using wavelet transform. and high-frequency detail coefficients of each layer Constructing an adaptive wavelet threshold The adaptive wavelet threshold The calculation formula includes:

[0038] in σ For noise variance, j The wavelet decomposition level; Based on an adaptive wavelet threshold, a hierarchical processing rule for wavelet coefficients is established, which includes high-frequency detail coefficients at each level. Output the correction coefficients after threshold processing based on the hierarchical processing rules. The correction coefficients and the low-frequency approximation coefficients are combined. Perform inverse wavelet transform reconstruction to obtain the second signal.

[0039] Furthermore, the step of formulating wavelet coefficient hierarchical processing rules based on adaptive wavelet thresholds, and processing the high-frequency detail coefficients of each layer... Output the correction coefficients after threshold processing based on the hierarchical processing rules. ,include: The wavelet coefficient hierarchical processing rules include:

[0040] In the formula, These are the corrected high-frequency coefficients after thresholding, used for signal reconstruction.

[0041] If the absolute value of the high-frequency detail factor satisfies Then the high-frequency detail coefficients will be... Set it directly to zero as the correction factor ; If the absolute value of the high-frequency detail factor satisfies Then the high-frequency detail coefficients will be... The correction factor is obtained by multiplying by a scaling factor and then compressing the product. ; If the absolute value of the high-frequency detail factor satisfies If a soft threshold shrinkage process is applied as a correction coefficient, then the soft threshold shrinkage process is performed through... Calculate the correction factor , where sgn( The parentheses are symbolic functions. If the value of the variable inside the parentheses is greater than 0, the function value is 1; if the value of the variable inside the parentheses is equal to 0, the function value is 0; if the value of the variable inside the parentheses is less than 0, the function value is -1.

[0042] The beneficial effects of the above steps are as follows: Unlike the traditional single hard / soft threshold, the three-stage hierarchical constraint can directly filter out small high-frequency background noise below half the threshold, compress and retain the critical weak components between half and full thresholds, and apply soft threshold shrinkage processing to the effective signal features above the threshold. While effectively filtering out invalid noise, it retains the sensitive features of impact faults in motor drive equipment to the maximum extent, taking into account both noise reduction and fault feature retention.

[0043] Further, in step 3, the normal threshold range of the time-domain amplitude of the second signal for each channel is calculated, and the abnormal type of each second signal is identified based on a pre-defined quantization classification and identification mechanism, including: The average value of the second signal amplitude on each channel is calculated as follows: μ and standard deviation σ And based on the mean and standard deviation, the normal threshold range of the second signal is calculated as [ , ],in , The 2.5σ optimized threshold is used, which has higher sensitivity than 3σ and lower false positives than 2σ. A pre-defined quantitative classification and identification mechanism is established, and the second signal is output as an anomaly type of random outlier, periodic electromagnetic interference, temperature drift and normal signal based on the pre-defined quantitative classification and identification mechanism. The pre-defined quantitative classification and identification mechanism includes an identification mechanism based on the duration of instantaneous anomalies, an identification mechanism based on periodic repetition, and an identification mechanism based on slowly varying drift characteristics, including: Acquire all sampling points on any channel within the same processing cycle and arrange them in chronological order of acquisition time. Record all sampling points whose signal amplitude exceeds the normal threshold range after arrangement as outliers, and count the number of outliers. ; The mechanism for identifying the duration of the instantaneous anomaly includes: if =1, and the second signal before and after the collection time of the anomaly point is within the normal threshold range, then the anomaly type of the signal is determined to be a random outlier. The identification mechanism based on periodic repetition includes: if Within the set L signal observation windows, all anomalies are counted. Anomalies with adjacent acquisition times are grouped together to form an anomaly signal segment. All anomaly signal segments are arranged in order of acquisition time. If the percentage of the relative deviation between the actual time interval between the characteristic extreme points of two adjacent anomaly data segments and the preset fixed theoretical period T is less than a preset percentage threshold, then the anomaly type of the signal is determined to be periodic electromagnetic interference. Here, the characteristic extreme point is a local maximum or local minimum value in the anomaly data segment, and the fixed theoretical period T is the interference main frequency extracted from the signal frequency domain. The reciprocal of; The identification mechanism based on slowly varying drift features includes: if If the value is 0, then starting from the second sampling point, calculate the difference between the temperature value of the current sampling point and the temperature value of the sampling point at the previous moment as the first-order difference of the current sampling point. If the first-order difference of 30 consecutive sampling points maintains the same sign, and the cumulative temperature change of 30 consecutive sampling points is greater than 0.5℃, then the abnormal type of the signal is determined to be temperature drift. Furthermore, the cumulative temperature change includes the absolute value of the temperature difference between the first and last sampling points out of all 30 sampling points.

[0044] Furthermore, second signals that do not conform to any identification mechanism are judged as normal signals, and the original normal signals are output as third signals.

[0045] Further, in step 4, the third signal is obtained by removing anomalies from the second signal of each anomaly type based on a pre-defined quantization anomaly removal mechanism, including: A pre-defined quantitative anomaly removal mechanism is used to remove the second signals of each anomaly type to obtain the third signal. The pre-defined quantitative anomaly removal mechanisms include: a precise removal mechanism based on random outliers, a precise removal mechanism based on periodic electromagnetic interference, and a precise removal mechanism based on temperature drift. A precise outlier removal mechanism based on random outliers: using the index k of outlier types as the center. To avoid smoothing out fault impact boundaries using ordinary mean interpolation, an adaptive window and inverse distance weighted correction are employed: Centered on the outlier index k, and the index k of the outlier of the random outlier type, two sampling points before and after the outlier in terms of collection time are selected as initial candidate points. If the number of valid normal points in the initial candidate points is insufficient, the search is automatically expanded to five points before and after, and the search range is further expanded to the preset maximum boundary until the preset number of valid candidate points is found or the upper limit of the search range is reached. Outliers that still exist in the initial candidate points are then removed, resulting in a set of valid candidate points Ω. The weight of each valid candidate point i in the set Ω is then calculated. Using the weighted average formula Calculate the corrected replacement value and replace the original signal amplitude at the anomaly point with the corrected replacement value; wherein, The signal amplitude at valid candidate point i, The time series distance between the outlier index and the candidate index; A precise elimination mechanism based on periodic electromagnetic interference: All anomalous signal segments of periodic interference type are acquired. Frequency domain components are analyzed using Fourier spectrum analysis. After eliminating the power frequency fundamental component, narrowband interference frequencies with amplitudes greater than three times the mean of the remaining spectrum are selected from the remaining spectrum. A second-order bandstop notch filter is designed based on the narrowband interference master frequency, with a stopband range of [missing information]. The abnormal signal segment of periodic electromagnetic interference is processed by narrowband filtering based on the second-order band-stop notch wave, and the signal after filtering out periodic interference is output as the third signal. Furthermore, when there are multiple interfering frequencies in the second signal, multiple band-stop notch filters are cascaded to filter out the interfering frequencies of each cycle in turn.

[0046] A precise rejection mechanism based on temperature drift is used: An interval consisting of anomalies exhibiting temperature drift over a continuous acquisition period is defined as the drift interval. Temperature values ​​of all anomalies within this drift interval are collected to form a temperature sequence. A quadratic polynomial least squares fitting is performed on this temperature sequence to obtain the slow drift trend term. The temperature value in the temperature series is subtracted from the slow drift trend term, and the calibrated temperature signal is output as the third signal. .

[0047] Further, in step 5, the Kalman filtering smoothing process with residual discrimination is performed on each third signal to output a smoothed signal. During the filtering process, the filtering parameters are adaptively updated based on the comparison result between the filtering residual and the preset threshold, including: Using a standard discrete linear Kalman filter recursive framework, the third signal is input into the recursive framework, and the observed signal at the current time k is... Once acquired, and before updating the observations at the current moment, the optimal estimate based on the previous moment is extracted from the standard recursive framework. The generated one-step state prediction value ; The observed signal and the predicted state value of the first step Difference to generate filter residuals Where H is the state matrix of the Kalman filter framework; when If the signal smoothing is successful, the optimal state estimate obtained after updating the observation at the current time is directly output. As the final smoothed signal; when When it is determined that there is residual strong interference, the Kalman filter model process noise covariance Q and observation noise covariance R are adaptively updated, and the filtering iteration is re-executed on the third signal to recalculate the filtering residual. Set a maximum number of iterations When the detected filter residual is greater than Enter adaptive update mode; increment the iteration counter by 1 after each iteration. If the number of iterations reaches Afterwards, the calculated filter residual was still greater than If the system determines that the current signal belongs to an extreme strong transient interference or a sensor hardware failure, it will abandon further iteration and force the output of the latest optimal estimate. X ( k | k As a smoothing result, a filter anomaly alarm flag is generated and saved simultaneously for the smoothed signal output at this time. The adaptive update of the Kalman filter model process includes the noise covariance Q and the observation noise covariance R, which include: Calculate the residual deviation factor β , ; The updated process noise covariance Q is: Q_{new} = β×Q_{current}, where Q_{new} is the updated process noise covariance and Q_{current} is the current process noise covariance; Using the updated Q_{new} and the current observation noise covariance R Re-execute the Kalman filter iteration and recalculate the Kalman gain. K ( k And update the optimal state estimate. X ( k | k Then, calculate the filter residual again. If the filter residual after the re-iteration is still greater than Then, the observation noise covariance R is also appropriately amplified to R_{new} = 1.2×R, and the Kalman filter iteration steps are re-executed with the updated R_{new} until the filter residual meets the convergence condition or reaches the maximum number of iterations.

[0048] In this embodiment, the method for constructing the discrete linear Kalman filter model is as follows: This invention only performs signal smoothing state estimation, without external control input, and omits control terms. The construction steps are as follows; 1) Core equation: Equations of state:

[0049] Observation equation:

[0050] for k The theoretical true state of the signal at any given moment; A The state transition matrix is ​​set to 1 by default, assuming the signal is locally stationary. For impactful, non-stationary, rapidly changing signals such as vibration and current, it is adaptively adjusted to 0.95 to adapt to the rapid dynamic changes of the signal. for k The theoretical true state of the signal at time -1; This is process noise, characterizing operating condition fluctuations, and is zero-mean Gaussian white noise. To enable sensors to acquire observation signals in real time; H This is the observation matrix, with a value of 1, representing that the observed values ​​directly correspond to the true state of the signal. To observe the noise and characterize the sensor acquisition interference, zero-mean Gaussian white noise was used.

[0051] 2) Kalman iteration formula:

[0052] for k The covariance of the prediction error at each step of time; for k Error covariance updated after filtering at time -1; Q The process noise covariance is determined by statistical analysis of variable operating condition tests on the motor drive system. For Kalman gain, dynamically balance the weights of predicted and observed values; R To observe the noise covariance, it was calibrated by steady-state no-load test; for k The optimal estimate after time-time filtering update; for k The predicted state value at a given time step; , They are respectively A and H The transpose of the matrix; For observation update k Time-major filtering error covariance; I It is an identity matrix.

[0053] Further, in step 6, the smoothing signal undergoes synchronous calibration to obtain a fourth signal, including: Zero-crossing feature points of the smoothed signal on the reference channel are extracted as reference timestamps. The actual timestamps of the smoothed signals on other channels within the same processing cycle are read, and the time difference Δ between the actual timestamps of the smoothed signals on other channels and the reference timestamps is calculated. t ; If Δ t If the value is greater than a preset threshold, the signal amplitude and actual timestamp of the smoothed signal adjacent to the acquisition time of the smoothed signal on the other channels are obtained. The theoretical amplitude of the signal at the reference timestamp is calculated and reconstructed using a linear interpolation formula. The theoretical amplitude is then used to replace the original sampled value of the smoothed signal on the other channels at the actual timestamp to achieve precise alignment. In this embodiment, when there is a large time deviation (Δt>Ti / 2) between the signal to be calibrated from other channels and the reference timestamp of the reference channel, linear interpolation is used to reconstruct the amplitude of the signal to be calibrated at the reference timestamp. The specific process is as follows: Let the base timestamp be t ref From the original sequence of the signal to be calibrated, find the time closest and located at... t ref Two adjacent actual sampling times before and after t 0 and t 1 (of which) t 0< t ref < t 1), and obtain the original signal amplitude corresponding to these two sampling times. X ( t 0) and X ( t 1).

[0054] Using a linear interpolation formula, the signal to be calibrated at the reference timestamp is calculated. t ref Theoretical amplitude at the location X ( t ref The calculation formula is as follows: X ( t ref )= X ( t 0)+( X ( t 1) X ( t 0))×( t ref t 0) / ( t 1 t 0) The calculated theoretical amplitude X ( t ref The value of the signal to be calibrated at this reference time is used to replace the original signal data, thus completing high-precision time synchronization alignment.

[0055] If Δ t If the value is not greater than a preset threshold, the actual timestamp of the smoothed signal on the other channels is directly and forcibly rewritten to the reference timestamp to complete the physical alignment of the time dimension. Furthermore, the reference channel is a current signal channel, the zero-crossing feature point is the zero-crossing feature point of the current signal, and the other channels include a vibration signal channel and a temperature signal channel.

[0056] Furthermore, the parameters for the signal processing after calibration and synchronization include: noise standard deviation. σ Interference detection threshold, periodic interference main frequency The filter coefficients and other parameters are stored in memory. When adaptively processing the signal in the next cycle, the updated noise standard deviation is used. σ The interference level determination range and process parameters for the next cycle are dynamically corrected to form a continuous iterative optimization closed loop, thereby improving the stability and accuracy of signal processing under complex working conditions in the long term.

[0057] Furthermore, the use of the updated noise standard deviation σ Dynamically adjust the interference level determination range and process parameters for the next cycle to form a continuous iterative optimization closed loop, including: Calculate the current periodic noise standard deviation σ new The standard deviation of the noise from the preset reference noise level σ base The deviation rate is used to dynamically amplify or reduce the strong interference threshold for the next cycle according to a preset adjustment coefficient. IF H and low interference threshold IF L This forms a closed-loop iterative optimization.

[0058] In this embodiment, the system pre-sets a reference noise standard deviation σ. base (Calibrated by steady-state no-load test). In each closed-loop iteration, obtain the current periodic noise standard deviation σ calculated in step 5. new ; Calculate the deviation rate between the current noise standard deviation and the reference noise standard deviation: ; Dynamically adjust the strong interference threshold for the next cycle using the deviation rate. IF H and low interference threshold IF L The formula is adjusted as follows: I FH_new = IF H ×(1+α·Δσ) I FL_new = IF L ×(1+α·Δσ) Wherein, α is a preset adjustment coefficient (with a value range of 0.1 to 0.3).

[0059] When environmental noise increases (σ newWhen the value increases, the increase is achieved by increasing the preset adjustment coefficient. IF H and IF L To avoid accidental triggering; When the environment improves (σ) new When the value decreases, the reduction is achieved by decreasing the preset adjustment coefficient. IF H and IF L This improves the sensitivity to early, minor faults.

[0060] Although the invention has been described in considerable detail and particularly with regard to several of the described embodiments, it is not intended to limit itself to any of these details or embodiments or any particular embodiment, thereby effectively covering the intended scope of the invention. Furthermore, the invention has been described above with respect to embodiments foreseeable by the inventors in order to provide a useful description, and non-substantial modifications to the invention that have not yet been foreseen may still represent equivalent modifications.

Claims

1. An adaptive processing method for multi-parameter monitoring signals of a motor drive system, characterized in that, The method includes: Step 1: Synchronously acquire the first signals of the three channels of the system, and calculate the interference factor of the first signal of each channel. IF Based on the configured interference determination threshold, according to the interference factor IF The interference level of the first signal in each channel is determined as strong interference, moderate interference, or weak interference. Step 2: Perform improved adaptive wavelet threshold denoising on the first signal with strong interference level to obtain the second signal; for the first signal with moderate interference level, skip the improved adaptive wavelet threshold denoising and use the original first signal as the second signal; for the first signal with weak interference level, skip steps 3 and 4 and use the original first signal as the third signal. Step 3: Calculate the normal threshold range based on the second signal of each channel, and identify the abnormal type of each second signal based on the pre-defined quantization classification and identification mechanism. The abnormal types include random outliers, periodic electromagnetic interference and temperature drift. Step 4: Based on the second signal of each anomaly type, perform anomaly removal on the second signal of each anomaly type according to the pre-defined quantification anomaly removal mechanism to obtain the third signal; Step 5: Perform Kalman filtering with residual discrimination on each third signal to smooth the signal and output a smoothed signal. During the filtering process, the filtering parameters are adaptively updated based on the comparison between the filtering residual and the preset threshold. Step 6: Perform synchronous calibration on each smoothed signal to obtain the fourth signal; Step 7: Feed back the filter residuals and filter parameters from Step 5 and the periodic interference main frequency extracted in Step 4 to Step 1 to dynamically update the interference judgment threshold for the next cycle, forming a closed-loop iterative optimization.

2. The processing method according to claim 1, characterized in that, In step 1, the interference factor of the first signal in each channel is calculated respectively. IF According to the interference factor IF The interference level of the first signal in each channel is determined as strong interference, moderate interference, or weak interference, including: In one processing cycle, the sensor acquires the sampled values ​​of each channel of the system based on a preset sampling frequency as the first signal, and calculates the interference factor based on the sampled values ​​of each channel, the signal mean, and the number of sampling points. IF ,include: In the formula, x ( n ) is the first single-channel signal n A momentary sample value, The average value of the signal amplitude. N This represents the number of signal sampling points. Preset strong interference threshold IF H and low interference threshold IF L And the interference factor calculated for each channel IF With respect to the preset strong interference threshold IF H and low interference threshold IF L Comparisons are made, and based on the comparison results, the interference level of the first signal in each channel is determined to be strong interference, moderate interference, or weak interference, including; if IF ≥ IF H The interference level of the first signal on the corresponding channel is determined to be strong interference. if IF L ≤ IF < IF H The interference level of the first signal on the corresponding channel is determined to be moderate interference. if IF < IF L The interference level of the first signal on the corresponding channel is determined to be weak interference.

3. The processing method according to claim 1, characterized in that, In step 2, the first signal, which has a strong interference level, is subjected to improved adaptive wavelet threshold denoising to obtain the second signal, which includes: The first signal, which has a strong interference level, is decomposed into low-frequency approximation coefficients using wavelet transform. and several layers of high-frequency detail coefficients Constructing an adaptive wavelet threshold The adaptive wavelet threshold The calculation formula includes: in σ For noise variance, j The wavelet decomposition level; Based on an adaptive wavelet threshold, a hierarchical processing rule for wavelet coefficients is established, which includes high-frequency detail coefficients at each level. Output the correction coefficients after threshold processing based on the hierarchical processing rules. The correction coefficients and the low-frequency approximation coefficients are combined. Perform inverse wavelet transform reconstruction to obtain the second signal.

4. The processing method according to claim 3, characterized in that, The adaptive wavelet threshold-based rule for hierarchical processing of wavelet coefficients is used to process the high-frequency detail coefficients of each layer. Output the correction coefficients after threshold processing based on the hierarchical processing rules. ,include: If the absolute value of the high-frequency detail factor satisfies Then the high-frequency detail coefficients will be... Set it directly to zero as the correction factor ; If the absolute value of the high-frequency detail factor satisfies Then the high-frequency detail coefficients will be... The correction factor is obtained by multiplying by a scaling factor and then compressing the product. ; If the absolute value of the high-frequency detail factor satisfies If a soft threshold shrinkage process is applied as a correction coefficient, then the soft threshold shrinkage process is performed through... Calculate the correction factor , where sgn(·) is a sign function. If the value of the variable inside the parentheses is greater than 0, the function value is 1; if the value of the variable inside the parentheses is equal to 0, the function value is 0; if the value of the variable inside the parentheses is less than 0, the function value is -1.

5. The processing method according to claim 1, characterized in that, In step 3, the normal threshold range of the time-domain amplitude of the second signal for each channel is calculated, and the abnormal type of each second signal is identified based on a pre-defined quantization classification and identification mechanism, including: Calculate the average amplitude of the second signal on each channel. μ and standard deviation σ And based on the mean and standard deviation, the normal threshold range of the second signal is calculated as [ , ],in , ; A pre-defined quantitative classification and identification mechanism is established, and the second signal is output as an anomaly type of random outlier, periodic electromagnetic interference, temperature drift and normal signal based on the pre-defined quantitative classification and identification mechanism. The pre-defined quantitative classification and identification mechanism includes an identification mechanism based on the duration of instantaneous anomalies, an identification mechanism based on periodic repetition, and an identification mechanism based on slowly varying drift characteristics, including: Acquire all sampling points of any channel within the same processing cycle, and arrange them in chronological order of acquisition time. Record all sampling points whose signal amplitude exceeds the normal threshold range after arrangement as outliers, and count the number of outliers. ; The mechanism for identifying the duration of the instantaneous anomaly includes: if =1, and the second signal before and after the collection time of the anomaly point is within the normal threshold range, then the anomaly type of the signal is determined to be a random outlier. The identification mechanism based on periodic repetition includes: if Within the set L signal observation windows, all anomalies are counted. Anomalies with adjacent acquisition times are grouped together to form an anomaly signal segment. All anomaly signal segments are arranged in order of acquisition time. If the percentage of the relative deviation between the actual time interval between the characteristic extreme points of two adjacent anomaly data segments and the preset fixed theoretical period T is less than a preset percentage threshold, then the anomaly type of the signal is determined to be periodic electromagnetic interference. Here, the characteristic extreme point is a local maximum or local minimum value in the anomaly data segment, and the fixed theoretical period T is the interference main frequency extracted from the signal frequency domain. The reciprocal of; The identification mechanism based on slowly varying drift features includes: if =0, then starting from the second sampling point, calculate the difference between the temperature value of the current sampling point and the temperature value of the second signal at the previous moment, and use this difference as the first-order difference of the current sampling point; if a preset number of consecutive... Nc The first-order differences of each sampling point all retain the same sign and are continuous. Nc The cumulative temperature change at each sampling point is greater than a preset temperature threshold. T diff If the signal is determined to be an anomaly of temperature drift, then the cumulative temperature change is... Nc The absolute value of the temperature difference between the first and last sampling points in a sampling point; If a second signal that does not conform to any identification mechanism is determined to be a normal signal, the original normal signal is output as the third signal.

6. The processing method according to claim 5, characterized in that, In step 4, the third signal is obtained by removing anomalies from the second signal of each anomaly type based on the pre-defined quantization anomaly removal mechanism, including: A pre-defined quantitative anomaly removal mechanism is used to remove the second signals of each anomaly type from the third signal based on the pre-defined quantitative anomaly removal mechanism. The pre-defined quantitative anomaly removal mechanism includes: a precise removal mechanism based on random outliers, a precise removal mechanism based on periodic electromagnetic interference, and a precise removal mechanism based on temperature drift. A precise outlier removal mechanism based on random outliers: Centering on the index k of an outlier of the random outlier type, several sampling points before and after the outlier's sampling time are simultaneously acquired as initial candidate points. Outliers present in all initial candidate points are removed, and a predetermined number of remaining candidate points are retained as valid candidate points, forming a valid candidate point set Ω. The weight of each valid candidate point in set Ω is then calculated. And through the weighted average formula Calculate the corrected replacement value and replace the signal amplitude at the original anomaly point with the corrected replacement value, wherein... Let be the signal amplitude of a valid candidate point, and i be the index of the valid candidate point. The time series distance between the outlier index and the candidate index; A precise elimination mechanism based on periodic electromagnetic interference: All anomalous signal segments of periodic interference type are acquired. Frequency domain components are analyzed using Fourier spectrum analysis. After eliminating the power frequency fundamental component, narrowband interference frequencies with amplitudes greater than three times the mean of the remaining spectrum are selected from the remaining spectrum. A second-order bandstop notch filter is designed based on the narrowband interference master frequency, with a stopband range of [missing information]. The abnormal signal segment of periodic electromagnetic interference is processed by narrowband filtering based on the second-order band-stop notch wave, and the signal after filtering out periodic interference is output as the third signal. A precise elimination mechanism based on temperature drift is as follows: A continuous sampling point interval identified as a temperature drift type is defined as a drift interval; the temperature values ​​of all sampling points within the drift interval are extracted to form a temperature sequence to be fitted; a quadratic polynomial least squares fitting is performed on the temperature sequence to be fitted to obtain a slow drift trend term; the slow drift trend term is subtracted from the temperature values ​​in the temperature sequence to be fitted to obtain a calibrated temperature signal, and the calibrated temperature signal is used as the third signal.

7. The processing method according to claim 1, characterized in that, In step 5, the Kalman filtering with residual discrimination is performed on each third signal to smooth the signal and output a smoothed signal. During the filtering process, the filtering parameters are adaptively updated based on the comparison between the filtering residual and a preset threshold, including: Using a standard discrete linear Kalman filter recursive framework, the third signal is input into the recursive framework, and the observed signal at the current time k is... Once acquired, and before updating the observations at the current moment, the optimal estimate based on the previous moment is extracted from the standard recursive framework. The generated one-step state prediction value ; The observed signal and the predicted state value of the first step Difference to generate filter residuals ,in, H The state matrix of the Kalman filter framework; when If the signal smoothing is successful, the optimal state estimate obtained after updating the observation at the current time is directly output. As the final smoothed signal; when When it is determined that there is residual strong interference, the Kalman filter model process noise covariance Q and observation noise covariance R are adaptively updated, and the filtering iteration is re-executed on the third signal to recalculate the filtering residual. Set a maximum number of iterations When the detected filter residual is greater than Enter adaptive update mode; increment the iteration counter by 1 after each iteration. If the number of iterations reaches Afterwards, the calculated filter residual was still greater than If the current signal is determined to be either an extremely strong transient interference or a sensor hardware failure, then the iteration should be abandoned and the latest optimal estimate should be forcibly output. X ( k | k As a smoothing result, a filter anomaly alarm flag is generated and saved simultaneously for the smoothed signal output at this time. The adaptive update of the Kalman filter model process includes the noise covariance Q and the observation noise covariance R, which include: Calculate the residual deviation factor β , ; The updated process noise covariance Q is: Q_{new} = β×Q_{current}, where Q_{new} is the updated process noise covariance and Q_{current} is the current process noise covariance; Using the updated Q_{new} and the current observation noise covariance R Re-execute the Kalman filter iteration and recalculate the Kalman gain. K ( k And update the optimal state estimate. X ( k | k Then, calculate the filter residual again. If the filter residual after the re-iteration is still greater than Then, the observation noise covariance R is also appropriately amplified to R_{new} = 1.2×R, and the Kalman filter iteration steps are re-executed with the updated R_{new} until the filter residual meets the convergence condition or reaches the maximum number of iterations.

8. The processing method according to claim 1, characterized in that, In step 6, the smoothing signal undergoes synchronous calibration to obtain a fourth signal, including: Zero-crossing feature points of the smoothed signal on the reference channel are extracted as reference timestamps. The actual timestamps of the smoothed signals on other channels within the same processing cycle are read, and the time difference Δ between the actual timestamps of the smoothed signals on other channels and the reference timestamps is calculated. t ; If Δ t If the value is greater than a preset threshold, the signal amplitude and actual timestamp of the smoothed signal adjacent to the acquisition time of the smoothed signal on the other channels are obtained. The theoretical amplitude of the signal at the reference timestamp is calculated and reconstructed using a linear interpolation formula. The theoretical amplitude is then used to replace the original sampled value of the smoothed signal on the other channels at the actual timestamp to achieve precise alignment. If Δ t If the time stamp is not greater than a preset threshold, the actual timestamp of the smoothed signal on the other channels is directly rewritten to the reference timestamp to complete the physical alignment of the time dimension. The reference channel is a current signal channel, the zero-crossing feature point is the zero-crossing feature point of the current signal, and the other channels include a vibration signal channel and a temperature signal channel.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the adaptive processing method for multi-parameter monitoring signals of a motor drive system as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The method includes a computer program that, when executed on a computer, causes the computer to perform an adaptive processing method for multi-parameter monitoring signals of a motor drive system as described in any one of claims 1-8.