Inductance winding equipment fault detection method and system based on multi-sensor fusion

Abstract

The invention relates to the technical field of fault diagnosis, in particular to an inductance winding equipment fault detection method and system based on multi-sensor fusion, and the method comprises the following steps: injecting a high-frequency sinusoidal voltage signal into a servo motor stator winding of inductance winding equipment, and synchronously collecting a motor end voltage, a three-phase current and a rotor position encoder signal; and constructing an original multi-dimensional signal matrix. According to the invention, the high-frequency sinusoidal voltage signal is actively injected into the stator winding of the servo motor of the inductance winding equipment, and the motor end voltage, the three-phase current and the rotor position encoder signal are synchronously acquired, so that the active excitation response capture of the operation state is realized, and compared with a passive monitoring mode, the signal controllability and the response accuracy are better; in an original multi-dimensional signal matrix, sequence calibration is carried out through a unified timestamp standard, time alignment is carried out on electrical signals and motion signals, and the collaboration and effective association of multi-source data are enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault diagnosis, in particular to an inductance winding device fault detection method and system based on multi-sensor fusion. BACKGROUND

[0002] The inductance winding device fault detection method is used for identifying and diagnosing inductance winding device operation abnormalities, aiming to discover potential fault risks inside the device in time through comprehensive collection and analysis of electrical parameters and mechanical states during device operation.

[0003] In the prior art, there is a lack of active excitation means in the data acquisition stage, resulting in weakening of fault components in the response signal, low signal-to-noise ratio, easy to be submerged by background noise in a high interference environment, and difficulty in identifying fine-grained fault features. At the same time, there is a lack of frequency domain analysis and modulation feature operation, and the bearing type micro-damage signal is difficult to present a typical frequency mode, which is easy to miss. Therefore, improvement is needed. SUMMARY

[0004] The purpose of the present application is to solve the shortcomings in the prior art, and to provide an inductance winding device fault detection method and system based on multi-sensor fusion.

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme, an inductance winding device fault detection method based on multi-sensor fusion, comprising the following steps: A high-frequency sinusoidal voltage signal is injected into the stator winding of the servo motor of the inductance winding device, the motor terminal voltage, three-phase current and rotor position encoder signal are synchronously collected, an original multi-dimensional signal matrix is constructed, and then the time alignment calibration of each signal sequence in the original multi-dimensional signal matrix is performed to obtain a multi-source synchronous electrical and motion signal set; According to the multi-source synchronous electrical and motion signal set, the high-frequency current response component is extracted, the high-frequency current response component and the injected high-frequency sine and cosine reference signals are calculated respectively, two product signals are obtained, the two product signals are low-pass filtered to obtain demodulation signals, and then the demodulation signals are subjected to Fourier transform to generate a bearing fault modulation feature spectrum; According to the multi-source synchronous electrical and motion signal set, the mechanical loss power time sequence is separated out, the spectral kurtosis of the mechanical loss power time sequence is calculated, and a friction state spectral kurtosis quantitative value is obtained; Based on the bearing fault modulation feature spectrum and the friction state spectrum kurtosis quantization value, the amplitude of the target fault feature frequency is extracted from the bearing fault modulation feature spectrum, and the friction health value is calculated. The fault feature frequency amplitude and the friction health value are combined into a winding equipment fusion state feature vector. Each element in the winding equipment fusion state feature vector is compared with the corresponding inductor winding equipment fault judgment threshold one by one to generate a comprehensive fault diagnosis conclusion for the inductor winding equipment.

[0006] Preferably, the step of acquiring the multi-source synchronous electrical and motion signal set is as follows: A high-frequency sinusoidal voltage signal is injected into the stator winding of the servo motor of the inductor winding equipment to trigger the synchronous sampling control pulses of the motor terminal voltage sensor, three-phase current sensor and rotor position encoder. The motor terminal voltage value, three-phase current value and rotor position encoder value of each sampling are recorded, and a timestamp generated by a unified reference clock is added to each record to form a synchronous sampling record with timestamp. Based on the timestamped synchronous sampling records, the motor terminal voltage values, three-phase current values, and rotor position encoder values ​​are sorted in columnar order according to the timestamps. The sampling sequence numbers within each column are unified, and the timestamp column is placed in the first column as a fixed alignment reference. The four columns are merged into a structured matrix to generate the original multidimensional signal matrix. Based on the original multidimensional signal matrix, the interval difference between adjacent entries in the timestamp column is calculated and the entries with abnormal intervals are marked. Missing entries are filled and duplicate entries are merged in the motor terminal voltage value column, the three-phase current value column and the rotor position encoder value column at the corresponding timestamp positions to form a multi-source synchronous electrical and motion signal set.

[0007] Preferably, the steps for obtaining the two product signals are as follows: Based on the multi-source synchronous electrical and motion signal set, the response range of the injected signal frequency in the three-phase current values ​​is located. The narrow-band components of the three-phase current values ​​in the response range are extracted in the order of timestamps. The fundamental component and the fixed frequency component associated with the operating condition are removed. The amplitude and phase information of the injected signal frequency band and its neighborhood are retained to obtain the high-frequency current response component. Based on the high-frequency current response component, the high-frequency current response component is multiplied by the injected high-frequency sinusoidal reference signal sample by sample and recorded as the first path according to the timestamp. The high-frequency current response component is multiplied by the injected high-frequency cosine reference signal sample by sample and recorded as the second path according to the timestamp. The first path and the second path are then combined in parallel according to the timestamp index to obtain two product signals.

[0008] Preferably, the step of obtaining the bearing fault modulation feature spectrum is as follows: Based on the two product signals, constraint parameters such as cutoff frequency, stopband attenuation, and transition bandwidth are set for each signal. Low-pass frequency band is extracted and the corresponding demodulated signal is output. The sampling length of the two demodulated signals is unified and frequency domain discretization and amplitude statistics are performed. The amplitude sequence is summarized according to the common frequency index to generate the bearing fault modulation characteristic spectrum.

[0009] Preferably, the step of obtaining the mechanical loss power time series is as follows: Based on the multi-source synchronous electrical and motion signal set, the three-phase current signal is transformed according to the sampling time to obtain the quadrature-axis current component and the direct-axis current component. The instantaneous electromagnetic torque is calculated by calling the known motor inductance and flux linkage parameters. At the same time, a dynamic observer is constructed based on the rotor position encoder signal to estimate the smooth rotor instantaneous angular velocity and angular acceleration. Then, the inertial torque is calculated based on the rotor rotational inertia. The inertial torque is subtracted from the instantaneous electromagnetic torque to obtain the total mechanical load torque time series. Based on the total mechanical load torque time series, polynomial fitting is performed within the sliding time window to decompose the torque time series into a smoothed baseline component representing the process load and a residual component representing friction and disturbance. The residual component is regarded as the instantaneous friction torque and multiplied point by point with the instantaneous angular velocity of the smoothed rotor to obtain the mechanical loss power time series. Outlier removal and boundary smoothing are performed on the mechanical loss power time series to generate the mechanical loss power time series.

[0010] Preferably, the step of obtaining the kurtosis quantification of the friction state spectrum is as follows: Based on the mechanical loss power time series, an enhanced sequence is first generated by applying an energy operator transformation to the sequence, and then a Fourier transform is performed on the enhanced sequence to obtain the spectral amplitude. The kurtosis quantization of the friction state spectrum is then calculated.

[0011] Preferably, the step of obtaining the fusion state feature vector of the winding device is as follows: Based on the bearing fault modulation characteristic spectrum, the target fault characteristic frequency point is located according to the frequency resolution. The amplitude of adjacent sampling points is interpolated according to the frequency difference ratio and spectral leakage is corrected to extract the target fault characteristic frequency amplitude. At the same time, it is matched with the friction state spectral kurtosis quantification value under the same timestamp to obtain the target fault characteristic frequency amplitude and friction state spectral kurtosis quantification value pairing entry. The friction health value is calculated by pairing the target fault characteristic frequency amplitude with the friction state spectrum kurtosis quantification value. The target fault characteristic frequency amplitude and the friction health value are combined in a fixed order to form a fusion state feature vector of the winding equipment.

[0012] Preferably, the steps for obtaining the comprehensive fault diagnosis conclusion of the inductor winding equipment are as follows: Each element in the fusion state feature vector of the winding equipment is compared with the fault judgment threshold of the inductor winding equipment, out-of-bounds elements are marked and the results are summarized to form a comprehensive fault diagnosis conclusion of the inductor winding equipment.

[0013] This invention also provides an inductor winding equipment fault detection system, comprising: The signal acquisition module is used to inject high-frequency sinusoidal voltage signals into the stator winding of the servo motor of the inductor winding equipment, and simultaneously acquire motor terminal voltage, three-phase current and rotor position encoder signals to construct an original multi-dimensional signal matrix. Then, the time alignment calibration of each signal sequence inside the original multi-dimensional signal matrix is ​​performed to obtain a multi-source synchronous electrical and motion signal set. The feature spectrum extraction module is used to extract the high-frequency current response component from the multi-source synchronous electrical and motion signal set, calculate the high-frequency current response component and the injected high-frequency sine and cosine reference signals respectively to obtain two product signals, perform low-pass filtering on the two product signals to obtain demodulated signals, and then perform Fourier transform on the demodulated signals to generate bearing fault modulation feature spectrum. The friction state analysis module is used to separate the mechanical loss power time series based on the multi-source synchronous electrical and motion signal set, calculate the spectral kurtosis of the mechanical loss power time series, and obtain the quantified value of the friction state spectral kurtosis. The comprehensive diagnostic module is used to extract the amplitude of the target fault characteristic frequency from the bearing fault modulation characteristic spectrum based on the bearing fault modulation characteristic spectrum and the friction state spectrum kurtosis quantification value, calculate the friction health value, combine the fault characteristic frequency amplitude and the friction health value into a winding equipment fusion state feature vector, and compare each element in the winding equipment fusion state feature vector with the corresponding inductor winding equipment fault judgment threshold one by one to generate a comprehensive fault diagnosis conclusion for the inductor winding equipment.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: This invention achieves active excitation response capture of the operating state by actively injecting a high-frequency sinusoidal voltage signal into the stator winding of the servo motor of an inductor winding device, and simultaneously acquiring the motor terminal voltage, three-phase current, and rotor position encoder signals. Compared with passive monitoring methods, this method offers greater signal controllability and response accuracy. In the original multi-dimensional signal matrix, sequence calibration is performed using a unified timestamp standard to align electrical and motion signals in time, enhancing the synergy and effective correlation of multi-source data. During fault feature extraction, the calculation and filtering of the high-frequency current response and high-frequency sine and cosine reference signals, combined with Fourier transform, accurately generates the modulation feature spectrum, enhancing the detection of weak axes. Building upon the identifiability of fault signals in the frequency domain, this paper further constructs a time series of mechanical loss power by input power and the rate of change of rotational energy. Combined with spectral kurtosis calculation, it achieves a quantitative assessment of frictional state changes. Thus, a frictional health value is constructed through an exponential mapping function, ensuring the continuity, stability, and boundedness of the health assessment and avoiding the risk of numerical overflow or misjudgment. Finally, by constructing a fused state feature vector from the target fault frequency amplitude and the frictional health value, and comparing it item by item with the fault judgment threshold, a multi-dimensional fusion diagnosis is achieved. This approach does not rely on a single indicator, improves the accuracy of diagnosis and anti-interference capability, and enhances the fault detection rate and maintenance response efficiency of inductor winding equipment in complex operating environments. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the steps of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0017] Please see Figure 1 This invention provides a technical solution: a fault detection method for inductor winding equipment based on multi-sensor fusion, comprising the following steps: A high-frequency sinusoidal voltage signal is injected into the stator winding of the servo motor of the inductor winding equipment. The motor terminal voltage, three-phase current and rotor position encoder signal are collected synchronously to construct the original multi-dimensional signal matrix. Then, the time alignment calibration of each signal sequence in the original multi-dimensional signal matrix is ​​performed to obtain a multi-source synchronous electrical and motion signal set. Based on the multi-source synchronous electrical and motion signal set, the high-frequency current response component is extracted. The high-frequency current response component is calculated with the injected high-frequency sine and cosine reference signals to obtain two product signals. The two product signals are then low-pass filtered to obtain the demodulated signal. Finally, the demodulated signal is Fourier transformed to generate the bearing fault modulation characteristic spectrum. Based on the multi-source synchronous electrical and motion signal set, the mechanical loss power time series is separated, the spectral kurtosis of the mechanical loss power time series is calculated, and the quantified value of the spectral kurtosis of the friction state is obtained. Based on the kurtosis quantization values ​​of the bearing fault modulation characteristic spectrum and the friction state spectrum, the amplitude of the target fault characteristic frequency is extracted from the bearing fault modulation characteristic spectrum, and the friction health value is calculated. The fault characteristic frequency amplitude and the friction health value are combined into a fusion state feature vector of the winding equipment. Each element in the fusion state feature vector of the winding equipment is compared with the corresponding fault judgment threshold of the inductor winding equipment one by one to generate a comprehensive fault diagnosis conclusion of the inductor winding equipment.

[0018] The steps for acquiring multi-source synchronous electrical and motion signal sets are as follows: A high-frequency sinusoidal voltage signal is injected into the stator winding of the servo motor of the inductor winding equipment to trigger the synchronous sampling control pulses of the motor terminal voltage sensor, three-phase current sensor and rotor position encoder. The motor terminal voltage value, three-phase current value and rotor position encoder value of each sampling are recorded, and a timestamp generated by a unified reference clock is added to each record to form a synchronous sampling record with timestamp. Based on the timestamped synchronous sampling records, the values ​​of motor terminal voltage, three-phase current and rotor position encoder are sorted in columnar order according to the timestamp. The sampling sequence number in the column is unified and the timestamp column is placed in the first column as a fixed alignment reference. The four columns are merged into a structured matrix to generate the original multidimensional signal matrix. Based on the original multidimensional signal matrix, the interval difference between adjacent entries in the timestamp column is calculated and the entries with abnormal intervals are marked. Missing entries are filled and duplicate entries are merged in the motor terminal voltage value column, the three-phase current value column and the rotor position encoder value column at the corresponding timestamp positions to form a multi-source synchronous electrical and motion signal set.

[0019] Specifically, a high-frequency sinusoidal voltage signal is injected into the stator winding of the servo motor in the inductor winding equipment. This is achieved through a programmable signal generator and power amplifier, generating a sinusoidal voltage signal with a frequency set between 500Hz and 2kHz and an amplitude set to 5% to 10% of the motor's rated voltage. This frequency is significantly higher than the fundamental operating frequency of the servo motor. This signal is then superimposed onto the stator winding's power supply line via the power amplifier. Subsequently, a field-programmable gate array (FPGA) is used as the master clock source to generate a 10kHz synchronous sampling control pulse. This pulse signal is then distributed in parallel to the analog-to-digital converter (ADC) trigger pin of the motor terminal voltage sensor, the ADC trigger pin of the three-phase current sensor, and the position data from the rotor position encoder. At the rising edge of each pulse, the voltage and current sensors immediately complete an analog-to-digital signal conversion, while the rotor position encoder latches its current absolute position count into the data register. A 64-bit high-precision counter inside the FPGA, driven by a 100MHz crystal oscillator, records the current count as a timestamp for each sampling pulse, achieving a resolution of 10 nanoseconds. Then, the 16-bit digital values ​​of the motor terminal voltage, the three 16-bit digital values ​​of the three-phase current, and the 32-bit count value of the rotor position encoder, along with the corresponding 64-bit timestamp, are packaged into a data frame. All data frames are continuously transmitted to a data acquisition buffer, forming a synchronous sampling record with timestamps.

[0020] Based on the timestamped synchronous sampling records, a data processing program is initiated. This program first reads all data frames from the data acquisition buffer and parses each data frame into independent data points. Each data point contains a timestamp, motor terminal voltage value, A-phase current value, B-phase current value, C-phase current value, and rotor position encoder value. Then, all data points are loaded into a list structure in memory, and a quicksort algorithm is called, using the timestamp value as the unique sort key, to sort all data points in ascending order, ensuring that all events are arranged in their physical time sequence. After sorting, the program creates a new two-dimensional data structure with columns defined as: the first column is the sampling sequence number, and the second column is the time... The data structure consists of three columns: the first column contains the timestamp, the second column contains the timestamp, the third column contains the motor terminal voltage value, the fourth column contains the three-phase current value (stored as a vector containing three sub-elements), and the fifth column contains the rotor-position encoder value. Then, the sorted list of data points is traversed, and an auto-incrementing integer sequence starting from 1 is generated as the sampling sequence number and filled into the first column. At the same time, the timestamp, voltage, three-phase current, and position value corresponding to each data point are sequentially filled into the second to fifth columns. The timestamp is fixed in the first column as the absolute reference for subsequent data alignment. Finally, this filled two-dimensional data structure is converted into a dense matrix. Each row of the matrix represents the state at a synchronous sampling moment, and each column represents a specific measurement dimension, generating the original multi-dimensional signal matrix.

[0021] Based on the original multidimensional signal matrix, the timestamp column is first processed by calculating the difference between adjacent timestamps row by row to obtain a time interval sequence. Then, the mean and standard deviation of this sequence are calculated, and anomaly thresholds are set based on these two statistics. Specifically, the upper threshold is set to the mean plus 1.5 times the standard deviation, and the lower threshold is set to the mean minus 1.5 times the standard deviation. For example, if the calculated average time interval is 100 microseconds and the standard deviation is 2 microseconds, then the upper threshold is... microseconds, lower threshold is In microseconds, the time interval sequence is iterated. When a time interval is greater than 103 microseconds, it is considered a missing data event. The number of missing sampling points is calculated based on the ratio of this interval to the average value. For example, if the interval is 201.5 microseconds, the number of missing points is... A new row is then inserted at that position. The timestamp of the new row is calculated by linear interpolation of the preceding and following timestamps. The corresponding motor terminal voltage value, three-phase current value, and rotor position encoder value in the new row are also calculated and filled based on the data of the two valid points before and after. If a certain time interval is less than 97 microseconds, it is determined to be a data duplication event. The average value of the motor terminal voltage value and three-phase current value in the two adjacent records is merged, and the rotor position encoder value is retained from the record of the later time point. Then the previous record is deleted. After processing all abnormal entries, a matrix with equal intervals and complete data on the time axis is obtained, forming a multi-source synchronous electrical and motion signal set.

[0022] The steps for obtaining the two product signals are as follows: Based on the multi-source synchronous electrical and motion signal set, the response range of the injected signal frequency in the three-phase current values ​​is located. The narrow-band components of the three-phase current values ​​in the response range are extracted in the order of timestamps. The fundamental component and the fixed frequency component associated with the operating condition are removed. The amplitude and phase information of the injected signal frequency band and its neighborhood are retained to obtain the high-frequency current response component. Based on the high-frequency current response component, the high-frequency current response component is multiplied by the injected high-frequency sinusoidal reference signal sample by sample and recorded by timestamp as the first path. The high-frequency current response component is multiplied by the injected high-frequency cosine reference signal sample by sample and recorded by timestamp as the second path. The first path and the second path are then merged in parallel according to the timestamp index to obtain the two product signals.

[0023] Specifically, based on the multi-source synchronous electrical and motion signal set, a Fast Fourier Transform (FFT) is first applied to the three-phase current numerical sequence of any one phase to convert it to the frequency domain, thereby locating the response range of the injected signal frequency. For example, if the injected high-frequency sinusoidal voltage signal frequency is 1kHz, then a response range bandwidth is set with 1kHz as the center frequency in the calculated current spectrum. This bandwidth needs to cover the modulation sideband caused by bearing failure. Based on the geometric parameters and speed range of the equipment bearing, the maximum fault characteristic frequency is estimated to be no more than 200Hz. Therefore, the response range is set to extend 250Hz on both sides of the center frequency, i.e., from 750Hz to 1250Hz. Subsequently, a digital bandpass filter is designed to intercept the narrowband component of this range. An 8th-order Butterworth filter is selected, with its passband range set to 750Hz to 1250Hz, passband ripple less than 0.5dB, and stopband attenuation greater than 60dB. Before bandpass filtering, strong interference signals need to be filtered out. Specifically, a notch filter with a center frequency of 50Hz and a bandwidth of 5Hz is designed to remove the fundamental frequency component. At the same time, based on the real-time speed calculated from the rotor position encoder signal of the multi-source synchronous electrical and motion signal set, a series of notch filters are dynamically set to remove fixed frequency components associated with the operating conditions. For example, when the motor speed is 3000 revolutions per minute (i.e., 50Hz), notch filters with a center frequency and a bandwidth of 5Hz are set at the fundamental frequency (50Hz, which coincides with the power frequency), the second harmonic (100Hz), and the third harmonic (150Hz) of the rotational frequency. The original three-phase current value sequence is processed by these notch filters and the Butterworth bandpass filter designed above. The final output signal sequence is the time sequence that only retains the amplitude and phase information of the injected signal frequency band and its neighborhood, thus obtaining the high-frequency current response component.

[0024] Based on the high-frequency current response components, corresponding high-frequency sinusoidal reference signals and high-frequency cosine reference signals of the same frequency and phase need to be generated. The frequencies of these two reference signals are exactly the same as the frequency of the high-frequency sinusoidal voltage signal injected into the stator winding of the servo motor, for example, 1kHz. Their phases are aligned with the initial phase of the injected signal, usually set to zero phase. Using the timestamp sequence from the multi-source synchronous electrical and motion signal set, precise instantaneous values ​​of the reference signals are generated for each sampling moment. Specifically, for any timestamp... The value of the high-frequency sinusoidal reference signal is calculated as follows: The value of the high-frequency cosine reference signal is calculated as follows: The amplitudes are all normalized to 1. Then, each data point of the high-frequency current response component is traversed, at each timestamp. The amplitude of the current response component at that moment. The product signal is multiplied by the high-frequency sine reference signal value and the high-frequency cosine reference signal value at the same time, respectively. The calculation process of the first product signal is as follows: and Multiply them and record the results in timestamp order. The calculation process for the second product signal is as follows: and The two signals are multiplied and the results are recorded in timestamp order. After the calculation of all sampling points is completed, two independent product signal time series are formed. Finally, a new data matrix is ​​created. The first column of the matrix is ​​the timestamp index, the second column stores the sequence of the first product signal, and the third column stores the sequence of the second product signal. The two signals are then merged in parallel to obtain two product signals.

[0025] The steps for obtaining the modulation characteristic spectrum of bearing faults are as follows: Based on the two product signals, constraint parameters such as cutoff frequency, stopband attenuation, and transition bandwidth are set for each signal. Low-pass frequency band is extracted and the corresponding demodulated signal is output. The sampling length of the two demodulated signals is unified and frequency domain discretization and amplitude statistics are performed. The amplitude sequence is summarized according to the common frequency index to generate the bearing fault modulation characteristic spectrum.

[0026] Specifically, based on the two product signals, a digital low-pass filter with identical parameters is designed and applied to each signal to extract the low-frequency modulated signal containing fault information. The filter parameter settings are crucial, and the specific constraints are as follows: the cutoff frequency is set based on the expected bearing fault characteristic frequency range. It is known that the bearing fault characteristic frequency of this equipment typically does not exceed 300Hz. To fully retain all potential fault information and provide a certain margin, the cutoff frequency is set to 400Hz. This value is much lower than the double carrier frequency component (i.e., 2kHz) generated during demodulation, effectively separating the modulated signal. The stopband attenuation is set to a large value, such as -80dB, to filter out high-frequency noise at and near the double carrier frequency. The transition bandwidth is set to 100Hz, meaning the filter's frequency response decreases rapidly between 400Hz and 500Hz. A finite pulse filter is designed using the window function method. Impulse Response (FIR) filters are used to meet the above parameters. A Kaiser window is selected, and the required filter order and coefficients are calculated. Then, the first and second product signals are convolved through this FIR low-pass filter. The output is the two demodulated signals. Before frequency domain analysis, the two demodulated signals are length-aligned, and 8192 consecutive data points are uniformly truncated. Then, Fast Fourier Transform (FFT) is performed on the two demodulated signals of the same length to obtain their respective complex spectra. In order to obtain accurate amplitude spectra that are not affected by phase, the two orthogonal spectral information needs to be merged. At each discrete frequency point, the square of the amplitude of the first spectrum is added to the square of the amplitude of the second spectrum, and the square root is taken as the final synthesized amplitude at that frequency point. The synthesized amplitudes of all frequency points are summarized with their corresponding frequency indices to generate the bearing fault modulation characteristic spectrum.

[0027] The steps for obtaining the time series of mechanical loss power are as follows: Based on the multi-source synchronous electrical and motion signal set, the three-phase current signal is transformed according to the sampling time to obtain the quadrature-axis current component and the direct-axis current component. The instantaneous electromagnetic torque is calculated by calling the known motor inductance and flux linkage parameters. At the same time, a dynamic observer is constructed based on the rotor position encoder signal to estimate the smooth rotor instantaneous angular velocity and angular acceleration. Then, the inertial torque is calculated based on the rotor rotational inertia. The inertial torque is subtracted from the instantaneous electromagnetic torque to obtain the total mechanical load torque time series. Based on the total mechanical load torque time series, polynomial fitting is performed within the sliding time window to decompose the torque time series into a smoothed baseline component representing the process load and a residual component representing friction and disturbance. The residual component is regarded as the instantaneous friction torque and multiplied point by point with the instantaneous angular velocity of the smoothed rotor to obtain the mechanical loss power time series. Outlier removal and boundary smoothing are performed on the mechanical loss power time series to generate the mechanical loss power time series.

[0028] Specifically, based on the multi-source synchronous electrical and motion signal set, three-phase current signals are extracted from it. , , and rotor position encoder signal, where the rotor position encoder signal provides the mechanical angle. First, the mechanical angle is converted into an electrical angle. The conversion relationship is as follows: ,in This is the number of pole pairs of the servo motor, a fixed parameter obtained from the motor's nameplate. For example, for an 8-pole motor... The value is 4. Subsequently, the Parker transform is applied to convert the current signal in the three-phase stationary coordinate system to the quadrature-axis current component in the two-phase rotating coordinate system. and direct-axis current components The specific calculation process of this transformation will not be elaborated here; its core is the use of electrical angle. As the phase reference for the transformation matrix, the motor inductance and flux linkage parameters, including the d-axis inductance, are then retrieved from the system parameter library and pre-measured using a motor test bench. q-axis inductance and permanent magnet magnetic flux For example, the parameters of a certain model of servo motor are: , , Calculate the instantaneous electromagnetic torque based on the electromagnetic torque formula of a permanent magnet synchronous motor. Meanwhile, in order to obtain smooth instantaneous rotor angular velocity and angular acceleration, a Luneburg dynamic observer is constructed, which is based on the mechanical motion equations of the motor. The rotor position encoder signal is used as the measurement input of the observer. The internal state variables of the observer are continuously corrected through a feedback correction mechanism, and the filtered and smoothed instantaneous rotor angular velocity is output. and angular acceleration Then, based on the rotor moment of inertia obtained from the motor datasheet... ,For example Calculate the inertial torque using the rotational form of Newton's second law. Finally, the calculated instantaneous electromagnetic torque Subtract inertial torque The total mechanical load torque time series is obtained.

[0029] Based on the total mechanical load torque time series, a local polynomial fitting method based on a sliding time window is used to separate the process load and friction components. A fixed-length time window is set, the length of which is determined according to the periodic characteristics of the winding process. For example, if a complete winding cycle is approximately 0.5 seconds and the sampling frequency is 10kHz, the window length can be set to 5000 data points. This window slides point by point along the total mechanical load torque time series. At each window position, a third-order polynomial fitting is performed on the data points within the window using the least squares method. The value of the fitted polynomial at the center point of the window is regarded as the smoothed baseline component at that moment. This component represents the relatively gradual change of the process load generated by the winding process itself. The value of each point in the original total mechanical load torque time series is subtracted from the value of its corresponding smoothed baseline component, and the resulting difference sequence is the residual component. This residual component mainly contains high-frequency torque fluctuations caused by equipment mechanical friction and unmodeled random disturbances, and is approximated as instantaneous friction torque. Then, this instantaneous friction torque time series is compared with the smoothed rotor instantaneous angular velocity obtained in the previous step. The time series is multiplied point by point to obtain the instantaneous mechanical loss power time series. Since noise and artifacts may be introduced during the calculation, the power series needs to be post-processed. First, outlier removal is performed. The mean and standard deviation of the data are calculated within a sliding window of 100 data points. Any data point that deviates from the mean by more than 3 times the standard deviation is identified as an outlier and replaced with the median of the data in the window. After outlier removal, a Savitzky-Gore smoothing filter with a window size of 51 points and a polynomial order of 3 is applied to the entire series for boundary smoothing, finally generating the mechanical loss power time series.

[0030] The steps for obtaining the kurtosis quantization of the friction state spectrum are as follows: Based on the mechanical loss power time series, an enhanced sequence is first generated by applying an energy operator transform to the sequence. Then, a Fourier transform is performed on the enhanced sequence to obtain the spectral amplitude. The kurtosis quantification of the friction state spectrum is calculated using the following formula: ; in, This represents the kurtosis quantization of the friction state spectrum. To enhance the complex amplitude of the sequence at the r-th frequency index after Fast Fourier Transform, This represents the total number of frequency indices. The Gaussian normal is set to a value of 1. To smooth the instantaneous angular velocity time series of the rotor, To smooth the normalized variance of angular velocity, This is the operating condition sensitivity adjustment factor, used to control the suppression strength of operating condition fluctuations on the calculation results.

[0031] Specifically, the formula for calculating the kurtosis quantification of friction state spectral values ​​quantifies the friction state of equipment by calculating the kurtosis after suppressing operating condition fluctuations. Its numerator is the standard calculation form of spectral kurtosis, which is sensitive to the impact characteristics of signals and can effectively detect transient impact signals caused by friction anomalies or bearing damage. Simultaneously, an operating condition fluctuation suppression term is introduced in the denominator. This term utilizes the normalized variance of the instantaneous angular velocity of the smoothed rotor. To quantify the stability of the winding equipment's operating conditions, when the equipment is under non-stationary operating conditions such as speed change, speed fluctuations increase. As the denominator increases, the denominator becomes larger, thus reducing the... This design effectively distinguishes between signal impacts caused by the deterioration of actual friction conditions and spurious impacts introduced by drastic changes in operating conditions, reducing the false alarm rate and making fault diagnosis results more reliable and stable under complex operating conditions such as variable speed and variable load. The introduction of this feature provides adjustability to the suppression intensity, which can be optimized according to the characteristics of specific equipment and operating conditions to achieve the best fault identification effect.

[0032] To enhance the complex amplitude of the sequence at the r-th frequency index after Fast Fourier Transform (FFT), the steps are as follows: First, apply the Tiger-Kaiser Energy Operator (TKEO) to the mechanical loss power time series generated in the previous step. For discrete mechanical loss power time series... The sequence after its energy operator transformation is the enhanced sequence. The calculation is as follows This operator amplifies the high-frequency transient impulse components in a signal. Subsequently, an enhanced sequence of length N (e.g., N=4096) is extracted and subjected to a Fast Fourier Transform (FFT) to obtain a spectral sequence composed of N complex numbers. That is, the r-th frequency point in the spectrum sequence ( A complex value (from 1 to N) that contains both the amplitude and phase information of the frequency component.

[0033] The total number of frequency indices depends on the length N of the data segment used to perform the Fast Fourier Transform. In practical calculations, since the input mechanical loss power time series is a real signal, its Fourier transform result has conjugate symmetry. Therefore, usually only the first half of the spectrum is used for analysis, i.e., from the DC component to the Nyquist frequency. In this case, the total number of frequency indices... Values For example, if the number of FFT calculation points N selected is 4096, then the number of effective frequency indices is... That is .

[0034] This is a Gaussian kurtosis constant. This parameter is set to ensure that the expected value of the calculated spectral kurtosis value for a pure Gaussian white noise signal is 0. By subtracting a kurtosis constant of 1, the spectral kurtosis baseline under healthy conditions (where the signal is mainly Gaussian noise) can be calibrated to near 0. This helps in setting the fault judgment threshold later. In this method, according to the standard definition of spectral kurtosis, this constant... The value is fixed at 1.

[0035] To smooth the normalized variance of the angular velocity, this parameter is used to quantify the fluctuation of the motor's operating speed. It is obtained by taking the smoothed rotor instantaneous angular velocity time series obtained in the previous step. In this process, the same data segment corresponding in time to the mechanical loss power time series used for spectrum calculation is extracted, and the average value of this velocity data segment is first calculated. and variance Then through the formula Calculations were performed; for example, 4096 velocity data points were collected within a certain time period, and the average value was calculated to be 150 rad / s, with a variance of 2.25. Then the normalized variance of the smoothed angular velocity corresponding to this working condition is: .

[0036] This is the operating condition sensitivity adjustment factor. This parameter controls the suppression strength of operating condition fluctuations on the final calculation results. It is an adjustment parameter that needs to be optimized based on experience or experimental data. The setting process is as follows: During the equipment commissioning phase, long-term operating data of the equipment under various typical operating conditions (including constant speed, acceleration, deceleration, load switching, etc.) are collected in both healthy and known early fault states. For each candidate... Values ​​(e.g., from 0.5 to 5.0, with a step size of 0.1) are used to calculate the values ​​for all data segments. Value, then evaluate in Under the given value, the two categories of samples are healthy and faulty. The separation of value distributions (e.g., the ratio of inter-class distance to intra-class variance can be used as an evaluation metric) should be selected to maximize the separation between the two classes of samples, while maintaining a healthy state. The value with the smallest fluctuation across different operating conditions The value was set as the final setting after analyzing experimental data from a certain type of winding equipment. The best performance is achieved with version 2.0.

[0037] Calculations based on parameters: Given an augmented sequence that has undergone FFT, The calculation results at each frequency point are simplified as follows: ; ; Meanwhile, the corresponding normalized variance of the smoothed angular velocity Gaussian normal quantity Operating condition sensitivity adjustment factor .

[0038] Substitute the above parameter values ​​into the formula: ; ; ; ; The results indicate that the calculated kurtosis quantification value of the friction state spectrum is 5.5557. This value is a quantitative assessment of the current friction state of the equipment. Since this value is greater than 0, it indicates the presence of non-Gaussian, impulsive components in the mechanical loss power signal. A pre-set fault judgment threshold is calculated by collecting a large amount of data during long-term operation of the equipment in a healthy state. The value is determined based on its statistical distribution (e.g., the mean plus three times the standard deviation). For example, if the threshold for a healthy state is set to 2.5, then the currently calculated result 5.5557 is greater than this threshold, which means that the friction state of the equipment is abnormal, and there may be premature bearing wear, poor lubrication, or other mechanical failures that can produce instantaneous impacts.

[0039] The steps for obtaining the fusion state feature vector of the winding equipment are as follows: Based on the bearing fault modulation characteristic spectrum, the target fault characteristic frequency point is located according to the frequency resolution. The amplitude of adjacent sampling points is interpolated according to the frequency difference ratio and spectral leakage is corrected to extract the target fault characteristic frequency amplitude. At the same time, it is matched with the friction state spectrum kurtosis quantification value under the same time stamp to obtain the matching entry of the target fault characteristic frequency amplitude and the friction state spectrum kurtosis quantification value. The friction health value is calculated by pairing the target fault characteristic frequency amplitude with the friction state spectrum kurtosis quantification value. The calculation formula is as follows: ; Where H is the friction health value, It is a health index decay factor and is greater than 0. The kurtosis quantization of the friction state spectrum in the entry for pairing the target fault characteristic frequency amplitude with the kurtosis quantization value of the friction state spectrum. This means that when the spectral kurtosis quantization value is less than 0, it takes the value 0; when it is greater than or equal to 0, it takes the original value. The target fault characteristic frequency amplitude and friction health value are combined in a fixed order to form the fusion state feature vector of the winding equipment.

[0040] Specifically, based on the bearing fault modulation characteristic spectrum, the theoretical fault characteristic frequencies of the monitored bearing must first be calculated. These frequencies are functions of motor speed and bearing geometric parameters, including the ball passing through the outer ring frequency, the ball passing through the inner ring frequency, the ball rotation frequency, and the cage frequency. Taking a specific bearing model 6205 as an example, its pitch circle diameter is 39.04 mm, the ball diameter is 7.94 mm, the number of balls is 9, and the contact angle is 0 degrees. When the instantaneous angular velocity of the smooth rotor obtained from the multi-source synchronous electrical and motion signal collection is 3000 rpm (i.e., the rotational frequency is 50 Hz), the fault characteristic frequency of the ball passing through the outer ring is calculated to be 160.7 Hz. These calculated frequency values ​​are the target fault characteristic frequencies. Next, these frequency points are located in the bearing fault modulation characteristic spectrum. Since the spectrum is discrete, its frequency resolution is determined by the sampling rate and the number of FFT points. For example, if the sampling rate is 10 kHz and the number of FFT points is 8192, then the frequency resolution is... This means that the theoretical frequency of 160.7Hz is likely to fall between two discrete frequency points, such as 160.1Hz and 161.32Hz. To obtain an accurate amplitude, linear interpolation is used to calculate the interpolated amplitude based on the distance ratio between 160.7Hz and these two frequency points. Simultaneously, to correct for spectral leakage caused by signal truncation, a Hanning window is applied to the signal before performing the FFT, and the interpolated amplitude is multiplied by a correction coefficient determined by the window function type (approximately 2.0 for the Hanning window). This allows for the extraction of a more accurate target fault characteristic. The amplitude of the characteristic frequency is obtained, for example, the amplitude of the outer ring fault characteristic frequency is 0.08. Then, this amplitude is paired with the friction state spectrum kilometry value calculated in the same data time period. Each bearing fault modulation characteristic spectrum and each friction state spectrum kilometry value originate from a data window with start and end timestamps. By matching these timestamps, the two are associated to form a record containing timestamp, target fault characteristic frequency name, corrected amplitude, and friction state spectrum kilometry value, thus obtaining a pairing entry of target fault characteristic frequency amplitude and friction state spectrum kilometry value.

[0041] formula: The above formula is used to quantify the kurtosis of the unbounded friction state spectrum. Mapped to a standardized frictional health value in the range (0, 1]. When the friction state of the equipment is ideal ( (approaching 0 or negative), health value A value close to 1 indicates a healthy state. As friction conditions deteriorate, such as wear or lubrication failure leading to increased impact signals, The value will increase positively, due to the effect of the exponential decay function, the health value It will rapidly decrease and approach 0, representing a fault state, and utilizes the non-linear characteristics of the exponential function to make the health value... Small positive changes are insensitive in the initial stages, but once... Beyond a certain range, health values ​​will decrease drastically, among which... The operation ensures that when the signal is smoother than Gaussian noise ( The health value will not exceed the upper limit of 1, making the evaluation benchmark of the health value uniform and stable.

[0042] The health index decay factor determines the rate at which the health value decreases as the spectral kurtosis quantification increases. Its value directly affects the sensitivity of fault diagnosis. The acquisition process is as follows: First, during the equipment debugging and baseline data acquisition phase, a large amount of data on the equipment operating in a confirmed healthy state is collected, and the corresponding kurtosis quantification values ​​of the friction state spectrum are calculated. A healthy sample set is formed, and statistical analysis is performed on this sample set to determine the spectral kurtosis threshold of a health state. This threshold is typically set as the mean of the healthy sample set plus three standard deviations. For example, by analyzing 1000 sets of health data, we obtain... The mean is 0.8 and the standard deviation is 0.5. Can be set to Next, we define the friction health value when the kurtosis quantification value of the device's friction state spectrum exactly reaches this threshold. This should be a preset critical health level, such as 0.5 (representing a sub-healthy or warning state). Based on this correspondence, the solution can be derived. The value is calculated using the following formula: Substitute the values, This calculation This value is then used in the calculation of all subsequent friction health values.

[0043] The parameter is the friction state spectrum kurtosis quantification value in the pairing entry of target fault characteristic frequency amplitude and friction state spectrum kurtosis quantification value. This parameter is a direct quantitative indicator of the equipment's friction state and is calculated through the previous steps. It reflects the intensity of the impact component in the mechanical loss power signal. The larger the value, the more severe the transient impact caused by friction inside the equipment, and the higher the probability of failure. This parameter is dimensionless, and its value is directly extracted from the pairing entry of target fault characteristic frequency amplitude and friction state spectrum kurtosis quantification value generated in the previous step. In this example, the friction state spectrum kurtosis quantification value calculated above is directly used, and its value is 5.5557.

[0044] Calculations based on parameters: Substitute the parameters obtained in the previous steps into the formula, where the health index decay factor is... Friction state spectrum kurtosis quantization value .

[0045] ; because ,therefore .

[0046] ; ; ; The result indicates that the current friction health value of the equipment is 0.1875, reflecting the health status of the equipment. A health value of 1 represents that the equipment is in the most ideal health state, while the closer the value is to 0, the worse the health status. The calculated value of 0.1875 is much smaller than the preset health threshold value of 0.5 and is very close to 0. This reveals that there is a serious abnormality in the friction system of the equipment, and it may be in a clear fault state. This friction health value will form a multi-dimensional state vector together with other fault characteristics (such as the frequency amplitude of bearing fault characteristics).

[0047] The steps for obtaining the comprehensive fault diagnosis conclusion of inductor winding equipment are as follows: Each element in the fusion state feature vector of the winding equipment is compared with the fault judgment threshold of the inductor winding equipment, out-of-bounds elements are marked and the results are summarized to form a comprehensive fault diagnosis conclusion for the inductor winding equipment.

[0048] Specifically, the target fault feature frequency amplitude extracted in the previous steps and the calculated friction health value are arranged in a predefined fixed order to form a multi-dimensional vector, namely the winding equipment fusion state feature vector. For example, if three types of faults—outer ring, inner ring, and ball bearings—are monitored, this vector can be defined as [outer ring fault feature frequency amplitude, inner ring fault feature frequency amplitude, ball bearing fault feature frequency amplitude, friction health value]. Using the values ​​from the previous example, a specific instance of this vector is [0.08, 0.02, 0.01, 0.1875], where the amplitudes of the inner ring and ball bearings are example values. Subsequently, each element of this vector is compared one by one with a preset, corresponding inductor winding equipment fault judgment threshold vector. This threshold vector is obtained by statistical analysis of a large number of collected feature vectors during the baseline period of healthy equipment operation. Specifically, for each feature element, its mean and standard deviation under healthy conditions are calculated, and the "mean + The alarm threshold is set to "6 times the standard deviation". For friction health values, the threshold is a fixed critical value, such as 0.5. Therefore, a possible threshold vector is [0.05, 0.04, 0.03, 0.5]. The comparison process is to compare the winding device's fusion state feature vector [0.08, 0.02, 0.01, 0.1875] with the threshold vector [0.05, 0.04, ...]. Comparing [0.03, 0.5], it was found that the first element 0.08 is greater than 0.05, and the fourth element 0.1875 is less than 0.5 (note that the smaller the health value, the worse it is). These two elements were marked as out of bounds. According to the preset diagnostic rule base, these marked results were logically combined to form the final diagnostic conclusion. For example, the rule base defines that "if the amplitude of the outer ring fault characteristic frequency is out of bounds and the friction health value is out of bounds, the diagnostic conclusion is: the outer ring of the bearing is severely worn, accompanied by deterioration of the lubrication condition". Other elements are not out of bounds, so the judgment of related faults is not triggered. Finally, all triggered rule results are summarized to form a comprehensive fault diagnosis conclusion for the inductor winding equipment.

[0049] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for fault detection of an inductance winding device based on multi-sensor fusion, characterized in that, The method comprises the following steps: A high-frequency sinusoidal voltage signal is injected into a servo motor stator winding of an inductive winding device, motor terminal voltage, three-phase current and rotor position encoder signals are synchronously collected, an original multi-dimensional signal matrix is constructed, and each signal sequence in the original multi-dimensional signal matrix is time-aligned and calibrated to obtain a multi-source synchronous electrical and motion signal set; According to the multi-source synchronous electrical and motion signal set, a high-frequency current response component is extracted, the high-frequency current response component and the injected high-frequency sine and cosine reference signals are calculated respectively to obtain two product signals, the two product signals are low-pass filtered to obtain demodulation signals, and the demodulation signals are subjected to Fourier transform to generate a bearing fault modulation characteristic spectrum; According to the multi-source synchronous electrical and motion signal set, a mechanical loss power time sequence is separated, and a spectrum kurtosis of the mechanical loss power time sequence is calculated to obtain a friction state spectrum kurtosis quantitative value; According to the bearing fault modulation characteristic spectrum and the friction state spectrum kurtosis quantitative value, an amplitude of a target fault characteristic frequency is extracted from the bearing fault modulation characteristic spectrum, a friction health value is calculated, the amplitude of the fault characteristic frequency and the friction health value are combined into a winding device fusion state feature vector, and each element in the winding device fusion state feature vector is compared with a corresponding inductive winding device fault judgment threshold to generate an inductive winding device comprehensive fault diagnosis conclusion.

2. The multi-sensor fusion based inductance wire winding apparatus fault detection method according to claim 1, characterized in that, The acquisition step of the multi-source synchronous electrical and motion signal set is: A high-frequency sinusoidal voltage signal is injected into a servo motor stator winding of an inductive winding device, a synchronous sampling control pulse of a motor terminal voltage sensor, a three-phase current sensor and a rotor position encoder is triggered, motor terminal voltage values, three-phase current values and rotor position encoder values of each sampling are recorded, a timestamp generated by a unified reference clock is attached to each record, and a timestamped synchronous sampling record is formed; According to the timestamped synchronous sampling record, the motor terminal voltage values, the three-phase current values and the rotor position encoder values are arranged in columns in timestamp order, the sampling sequence numbers in the columns are unified, and the timestamp column is placed in the first column as a fixed alignment reference, four columns are combined into a structured matrix, and an original multi-dimensional signal matrix is generated; According to the original multi-dimensional signal matrix, interval differences between adjacent items in the timestamp column are calculated and interval abnormal items are marked, missing items in the motor terminal voltage value column, the three-phase current value column and the rotor position encoder value column are filled and repeated items are combined at corresponding timestamp positions, and a multi-source synchronous electrical and motion signal set is formed.

3. The multi-sensor fusion based inductance wire winding apparatus fault detection method according to claim 1, characterized in that, The acquisition step of the two product signals is: According to the multi-source synchronous electrical and motion signal set, a response interval of the injected signal frequency in the three-phase current values is located, narrow-band components of the three-phase current values in the response interval are intercepted in timestamp order, fundamental component and fixed frequency components associated with working conditions are removed, amplitude and phase information of the injected signal frequency band and neighborhood are retained, and a high-frequency current response component is obtained; According to the high-frequency current response component, the high-frequency current response component is multiplied with the injected high-frequency sine reference signal sample by sample and is time-stamped as a first path, the high-frequency current response component is multiplied with the injected high-frequency cosine reference signal sample by sample and is time-stamped as a second path, the first path and the second path are parallel-merged according to time-stamp indexes, and two-path product signals are obtained.

4. The multi-sensor fusion based inductive wire winding apparatus fault detection method according to claim 1, wherein, The bearing fault modulation feature spectrum acquisition step is: According to the two-path product signals, constraint parameters of a cut-off frequency, a stopband attenuation and a transition bandwidth are set for each path, low-frequency band interception is performed, and corresponding demodulation signals are output, the two-path demodulation signals are uniformly sampled in length, frequency domain discrete expansion and amplitude statistics are performed, amplitude sequences are summarized according to common frequency indexes, and a bearing fault modulation feature spectrum is generated.

5. The multi-sensor fusion based inductive wire winding apparatus fault detection method according to claim 1, wherein, The mechanical loss power time sequence acquisition step is: According to the multi-source synchronous electrical and motion signal set, three-phase current signals are coordinate-transformed according to sampling time to obtain a quadrature-axis current component and a direct-axis current component, an instantaneous electromagnetic torque is calculated by calling known motor inductance and flux linkage parameters, a dynamic observer is constructed according to a rotor position encoder signal to estimate smooth rotor instantaneous angular velocity and angular acceleration, an inertial torque is calculated according to rotor moment of inertia, and the instantaneous electromagnetic torque is subtracted from the inertial torque to obtain a total mechanical load torque time sequence; According to the total mechanical load torque time sequence, polynomial fitting is performed in a sliding time window, the torque time sequence is decomposed into a smooth baseline component representing a process load and a residual component representing friction and disturbance, the residual component is regarded as an instantaneous friction torque, and the instantaneous friction torque is multiplied point by point with the smooth rotor instantaneous angular velocity to obtain a mechanical loss power time sequence, and the mechanical loss power time sequence is subjected to outlier rejection and boundary smoothing processing to generate a mechanical loss power time sequence.

6. The multi-sensor fusion based inductive wire winding apparatus fault detection method as claimed in claim 1, wherein, The friction state spectrum kurtosis quantitative value acquisition step is: According to the mechanical loss power time sequence, an energy operator transformation is first applied to the sequence to generate an enhanced sequence, and then a Fourier transform is performed on the enhanced sequence to obtain a frequency spectrum amplitude, and a friction state spectrum kurtosis quantitative value is calculated.

7. The multi-sensor fusion based inductive wire winding apparatus fault detection method as claimed in claim 1, wherein, The winding equipment fusion state feature vector acquisition step is: According to the bearing fault modulation feature spectrum, a target fault feature frequency point is located according to a frequency resolution, adjacent sampling point amplitudes are interpolated according to a frequency difference ratio and are subjected to spectrum leakage correction, a target fault feature frequency amplitude is extracted, and the target fault feature frequency amplitude is correspondingly paired with a friction state spectrum kurtosis quantitative value at a unified time stamp to obtain a target fault feature frequency amplitude and friction state spectrum kurtosis quantitative value pairing item; According to the target fault feature frequency amplitude and friction state spectrum kurtosis quantitative value pairing item, a friction health value is calculated; The target fault feature frequency amplitude and the friction health value are taken in a fixed order to form a winding equipment fusion state feature vector.

8. The multi-sensor fusion based inductive wire winding apparatus fault detection method as claimed in claim 1, wherein, The inductance winding equipment comprehensive fault diagnosis conclusion acquisition step is: Each element in the winding equipment fusion state feature vector is compared with an inductance winding equipment fault judgment threshold value one by one, out-of-bound elements are marked, and results are summarized to form an inductance winding equipment comprehensive fault diagnosis conclusion.

9. The inductance winding apparatus fault detection system based on multi-sensor fusion according to any one of claims 1-8, characterized in that, It comprises: The signal acquisition module is configured to inject a high-frequency sinusoidal voltage signal into a stator winding of a servo motor of the inductive winding device, synchronously collect motor terminal voltage, three-phase current and rotor position encoder signals, construct an original multi-dimensional signal matrix, and then perform time alignment calibration on each signal sequence in the original multi-dimensional signal matrix to obtain a multi-source synchronous electrical and motion signal set. The characteristic frequency spectrum extraction module is configured to extract a high-frequency current response component from the multi-source synchronous electrical and motion signal set, calculate the high-frequency current response component and an injected high-frequency sine and cosine reference signal respectively, obtain two product signals, perform low-pass filtering on the two product signals to obtain demodulation signals, perform Fourier transform on the demodulation signals to generate a bearing fault modulation characteristic frequency spectrum, and extract a target fault characteristic frequency amplitude from the bearing fault modulation characteristic frequency spectrum. The friction state analysis module is configured to separate a mechanical loss power time sequence from the multi-source synchronous electrical and motion signal set, calculate a spectral kurtosis of the mechanical loss power time sequence, and obtain a friction state spectral kurtosis quantitative value. The comprehensive diagnosis module is configured to extract a target fault characteristic frequency amplitude from the bearing fault modulation characteristic frequency spectrum and calculate a friction health value according to the bearing fault modulation characteristic frequency spectrum and the friction state spectral kurtosis quantitative value, combine the fault characteristic frequency amplitude and the friction health value into a winding device fusion state feature vector, compare each element in the winding device fusion state feature vector with a corresponding inductive winding device fault judgment threshold one by one, and generate an inductive winding device comprehensive fault diagnosis conclusion.

Images