A monitoring and diagnosis system for rolling mill reducers based on multi-source information fusion

Abstract

The present invention relates to the technical field of measuring electrical variables, specifically a rolling mill reducer monitoring and diagnosis system based on multi-source information fusion, the system comprising: an electrical parameter signal acquisition module, which obtains continuous readings of the three-phase voltage and the three-phase current of the rolling mill drive motor, and synchronizes the readings. In the present invention, by synchronously acquiring the three-phase voltage and current data of the rolling mill drive motor and calibrating the amplitude and phase, a more accurate calibration electrical measurement set is obtained, effectively avoiding error propagation in the signal acquisition stage; the calibrated electrical data is used to directly calculate the instantaneous input electrical power, and after deducting the motor loss, an accurate motor shaft power sequence is obtained, and then the Fourier transform method is used to extract the key harmonic features in the shaft power spectrum; at the same time, the motor stator current is band-pass filtered and Hilbert transform analyzed to obtain a clearer and more recognizable spectrum kurtosis value, thereby enhancing the ability to determine current impact and periodic characteristics.

Description

Technical Field

[0001] The present invention relates to the technical field of measuring electrical variables, and in particular to a rolling mill reducer monitoring and diagnosis system based on multi-source information fusion. Background Art

[0002] The field of electrical variable measurement technology involves electrical engineering and automation monitoring technology, including real-time measurement and analysis of various electrical parameters such as voltage, current, power, frequency, and phase. Its application goal is to perceive the various dynamic characteristics and state changes during the operation of electrical equipment, and support fault diagnosis, operating status assessment, and equipment protection.

[0003] Existing technologies are susceptible to noise and phase errors, leading to significant errors in analysis results. Furthermore, condition assessments are based solely on a single electrical parameter or a few indicators, ignoring the inherent surge characteristics and periodic fluctuations in the current. This results in insufficient ability to identify latent equipment faults at an early stage, potentially delaying their discovery. Therefore, improvements are needed. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a rolling mill reducer monitoring and diagnosis system based on multi-source information fusion.

[0005] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: A rolling mill reducer monitoring and diagnosis system based on multi-source information fusion includes:

[0006] The electrical parameter signal acquisition module obtains continuous readings of the three-phase voltage and three-phase current of the rolling mill drive motor, synchronizes the readings, and performs amplitude and phase calibration to obtain a calibrated electrical measurement set;

[0007] a shaft power characteristic calculation module, calling the three-phase voltage and the three-phase current of the rolling mill drive motor in the calibration electrical measurement set, calculating the instantaneous input electric power, obtaining instantaneous electric power data, estimating the motor loss, and deducting the motor loss from the instantaneous electric power data to obtain an instantaneous motor shaft power sequence, performing a Fourier transform on the instantaneous motor shaft power sequence, and establishing key harmonics of the shaft power spectrum;

[0008] a current modulation analysis module, which extracts a rolling mill drive motor stator current signal from the calibration electrical measurement set, performs bandpass filtering on the rolling mill drive motor stator current signal to obtain a current in a frequency band of interest, applies a Hilbert transform to the current in the frequency band of interest to extract an envelope to obtain a frequency-selective current envelope sequence, calculates a spectral kurtosis value of the frequency-selective current envelope sequence, and obtains current impulse and period characteristics;

[0009] The reducer state determination module integrates the amplitude and phase information of the key harmonics of the shaft power spectrum and the kurtosis value and cyclic frequency characteristics of the current impact and periodic characteristics to construct a multidimensional electrical state vector, compares each element in the multidimensional electrical state vector with a preset benchmark item by item, and outputs a rolling mill reducer operating condition determination.

[0010] Preferably, the steps of obtaining the calibration electrical measurement set are:

[0011] Through the three-phase voltage sensor and the three-phase current sensor, the continuous readings of the three-phase voltage and the three-phase current of the rolling mill drive motor are obtained in real time, and the acquisition timestamp of each reading is recorded to form the original voltage and current data;

[0012] Based on the raw voltage and current data, the continuous readings of the three-phase voltage of the rolling mill drive motor and the continuous readings of the three-phase current of the rolling mill drive motor are matched one by one according to the acquisition timestamp, the time base is unified and the data are aligned to generate synchronized voltage and current data;

[0013] Based on the synchronized voltage and current data, the standard reference voltage signal and the standard reference current signal are called, and the amplitude comparison and phase difference compensation are performed on the synchronized voltage and current data item by item, and the voltage amplitude and current amplitude as well as the voltage phase and current phase errors are corrected to obtain a calibrated electrical measurement set.

[0014] Preferably, the steps for obtaining the instantaneous motor shaft power sequence are:

[0015] Extracting the three-phase instantaneous voltage and current values ​​of the rolling mill drive motor based on the calibration electrical measurement set, combining the data of each phase one by one according to the phase synchronization sampling time, recording the three sets of voltage and current pairs at each sampling time in the combination, and generating a three-phase instantaneous voltage and current combination sequence;

[0016] Calculating the total instantaneous input electric power at each sampling moment according to the three-phase instantaneous voltage and current combination sequence;

[0017] Based on the total instantaneous input electric power at each sampling moment, the calibrated efficiency curve data of the rolling mill drive motor at different load rates is called, the corresponding sampled power value is looked up in the table to obtain the loss power value, and the loss power value is subtracted from the total instantaneous input electric power item by item to generate the instantaneous motor shaft power sequence.

[0018] Preferably, the steps for obtaining the key harmonics of the shaft power spectrum are:

[0019] Based on the instantaneous motor shaft power sequence, continuous power value segments in the sequence are intercepted at fixed time intervals, and the time starting point and ending point of each segment are recorded to form a periodically segmented instantaneous shaft power segment sequence;

[0020] According to the instantaneous shaft power segment sequences of the periodic segments, Fourier transform is performed on each shaft power segment sequence one by one, the shaft power value is mapped from the time domain to the frequency domain, and frequency domain power spectrum data of each shaft power segment sequence is generated;

[0021] Based on the frequency domain power spectrum data, the defined target frequency points are retrieved, the amplitude value and the phase value corresponding to each target frequency point are respectively extracted from the frequency domain power spectrum data, and the key harmonics of the axis power spectrum are established.

[0022] Preferably, the step of obtaining the current in the frequency band of interest is:

[0023] Read the amplitude of each sampling point of the stator current signal of the rolling mill drive motor from the calibration electrical measurement set, arrange them in chronological order, intercept the current amplitudes of all sampling points in the most recent 30-second period to form a sampling segment, extract the maximum amplitude, average amplitude, total number of sampling points and fixed sampling interval value in the signal segment, and generate a time-limited stator current amplitude sequence;

[0024] Calculating a filter center frequency determination value based on the stator current amplitude sequence during the time limit period;

[0025] Based on the filtering center frequency judgment value, the upper and lower cutoff frequencies of the bandpass filter are set with the filtering center frequency judgment value as the center, the cutoff range is set using a fixed bandwidth mode, and the filter is executed to perform bandpass processing on the frequency spectrum of the stator current amplitude sequence in the time period limit to form a frequency band current of interest.

[0026] Preferably, the steps of obtaining the current impact and period characteristics are:

[0027] Extracting all sampling point data from the current in the frequency band of interest, performing Hilbert transform point by point and modulo the results to obtain the current envelope amplitude corresponding to each sampling point, forming a frequency-selective current envelope sequence, and extracting the maximum envelope amplitude, average amplitude, and original amplitude of each sampling point to form a normalized envelope amplitude structure;

[0028] Calculating a spectrum kurtosis value based on the normalized envelope amplitude structure;

[0029] Based on the spectrum kurtosis value, it is determined whether it is higher than the periodic impact recognition threshold. If it is higher than the periodic impact recognition threshold, it is determined that a high-amplitude impact feature exists. The periodic repeatability is then determined in combination with the uniformity of the time intervals between consecutive local peaks, and the current impact and periodic characteristics are output.

[0030] Preferably, the steps of obtaining the multi-dimensional electrical state vector are:

[0031] Calling the key harmonics of the shaft power spectrum, respectively extracting the harmonic amplitude value and the harmonic phase value corresponding to each target frequency point in the key harmonics of the shaft power spectrum, and pairing them one by one in the order of the frequency points to generate shaft power spectrum feature data including amplitude and phase;

[0032] Based on the shaft power spectrum characteristic data, the current impact and period characteristics are called, the spectrum kurtosis value and the cycle frequency characteristic data in the current impact and period characteristics are extracted one by one, and time synchronization is performed with the shaft power spectrum characteristic data according to the corresponding time to generate synchronized electrical characteristic data;

[0033] Based on the synchronized electrical characteristic data, the harmonic amplitude, harmonic phase, spectrum kurtosis value and cyclic frequency characteristics are arranged in order, and a characteristic parameter vector with time series correlation is constructed by combining them one by one to obtain a multidimensional electrical state vector.

[0034] Preferably, the steps for obtaining the working condition determination of the rolling mill reducer are:

[0035] Extracting the value of each characteristic parameter in the multidimensional electrical state vector, respectively calling the preset reference threshold of the corresponding characteristic parameter, calculating the numerical difference between each characteristic parameter and the corresponding reference threshold one by one, and generating a difference numerical sequence between the multidimensional electrical state vector and the reference threshold;

[0036] Based on the difference value sequence between the multidimensional electrical state vector and the reference threshold, determining whether the absolute value of each difference value exceeds the set fluctuation amplitude threshold one by one, recording the number and category of characteristic parameters exceeding the fluctuation amplitude threshold, and generating statistical results of the fluctuation amplitude exceeding characteristic parameters;

[0037] Based on the statistical results of the fluctuation amplitude exceeding limit characteristic parameters, the number and type of the exceeding limit characteristic parameters in the statistical results are analyzed. If the number of exceeding limit characteristic parameters exceeds the preset warning threshold and the type includes periodic characteristics or harmonic characteristics, it is determined that the operating condition of the rolling mill reducer is abnormal; otherwise, the operating condition is determined to be normal, thereby forming a rolling mill reducer operating condition judgment.

[0038] Compared with the prior art, the advantages and positive effects of the present invention are:

[0039] In the present invention, by synchronously collecting the three-phase voltage and current data of the rolling mill drive motor and calibrating the amplitude and phase, a more accurate calibrated electrical measurement set is obtained, which effectively avoids error propagation in the signal acquisition stage; the calibrated electrical data is used to directly calculate the instantaneous input electric power, and after deducting the motor loss, an accurate motor shaft power sequence is obtained, and then the Fourier transform method is used to extract the key harmonic features in the shaft power spectrum; at the same time, the motor stator current is band-pass filtered and Hilbert transform analyzed to obtain a clearer and more recognizable spectrum kurtosis value, thereby enhancing the ability to determine the current impact and periodic characteristics; further, the key harmonics of the shaft power spectrum are combined with the current impact and periodic characteristics to construct a multi-dimensional electrical state vector, which is compared with the preset benchmark item by item, so that the working condition determination of the rolling mill reducer is improved, early fault identification and operation risk warning are achieved, and the safety and stability of the rolling mill reducer are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present 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 only used to explain the present invention and are not intended to limit the present invention.

[0042] See also Figure 1 The present invention provides a technical solution: a rolling mill reducer monitoring and diagnosis system based on multi-source information fusion, comprising:

[0043] The electrical parameter signal acquisition module obtains continuous readings of the three-phase voltage and three-phase current of the rolling mill drive motor, synchronizes the readings, and performs amplitude and phase calibration to obtain a calibrated electrical measurement set;

[0044] The shaft power characteristic calculation module calls the three-phase voltage and current of the rolling mill drive motor in the calibration electrical measurement set, calculates the instantaneous input electric power, obtains instantaneous electric power data, estimates the motor loss, and deducts the motor loss from the instantaneous electric power data to obtain the instantaneous motor shaft power sequence. It performs Fourier transform on the instantaneous motor shaft power sequence and establishes the key harmonics of the shaft power spectrum.

[0045] The current modulation analysis module extracts the stator current signal of the rolling mill drive motor from the calibrated electrical measurement set, performs bandpass filtering on the stator current signal of the rolling mill drive motor to obtain the current in the frequency band of interest, applies the Hilbert transform to extract the envelope of the current in the frequency band of interest to obtain a frequency-selective current envelope sequence, calculates the spectral kurtosis value of the frequency-selective current envelope sequence, and obtains the current impulse and periodic characteristics;

[0046] The reducer state determination module integrates the amplitude and phase information of the key harmonics of the shaft power spectrum as well as the kurtosis value and cyclic frequency characteristics of the current impact and periodic characteristics to construct a multi-dimensional electrical state vector. It compares each element in the multi-dimensional electrical state vector with the preset benchmark item by item and outputs the working condition determination of the rolling mill reducer.

[0047] The steps to obtain a calibrated electrical measurement set are:

[0048] Through the three-phase voltage sensor and the three-phase current sensor, the continuous readings of the three-phase voltage and the three-phase current of the rolling mill drive motor are obtained in real time, and the acquisition timestamp of each reading is recorded to form the original voltage and current data;

[0049] Based on the original voltage and current data, the continuous readings of the three-phase voltage and the three-phase current of the rolling mill drive motor are matched one by one according to the acquisition timestamp, the time base is unified and the data is aligned to generate synchronized voltage and current data;

[0050] Based on the synchronized voltage and current data, the standard reference voltage signal and the standard reference current signal are called, and the amplitude comparison and phase difference compensation are performed on the synchronized voltage and current data item by item. The voltage amplitude and current amplitude as well as the voltage phase and current phase errors are corrected to obtain a calibrated electrical measurement set.

[0051] Specifically, through three-phase voltage sensors, such as voltage sensor modules, and three-phase current sensors, the sampling frequency of the data acquisition system is set for equipment such as rolling mill drive motors that usually have a rated power of hundreds to thousands of volt-amperes. The setting of the sampling frequency is based on the Nyquist sampling theorem and must be at least twice the highest frequency of interest (for example, analysis up to the 25th harmonic, i.e. 1250Hz or 1500Hz) in the harmonics of the motor power frequency (such as 50Hz or 60Hz), and considering the influence of the switching frequency of the switching device (such as IGBT), it is usually set in the range of 10kHz to 50kHz. For example, a sampling frequency of 20kHz is selected to ensure that sufficient signal details are captured and the rolling mill is obtained in real time. During the continuous reading of the three-phase (A, B, C) voltage and the three-phase current of the drive motor, the instantaneous analog quantity collected for each group (UA, UB, UC, IA, IB, IC) is converted into a digital quantity by its own independent analog-to-digital converter (ADC), and a timestamp is immediately added. The timestamp comes from a unified clock source synchronized with the main control system, such as a system clock synchronized to microsecond accuracy through the Network Time Protocol (NTP). The timestamp corresponding to each reading is recorded. These data streams containing voltage, current and their corresponding timestamps are initially aggregated and formatted to form structured raw voltage and current data. The data is stored in a sequence, and each element contains (timestamp, , , , , , ).

[0052] Based on the structured raw voltage and current data obtained in the previous step, which contains multiple time series that are not guaranteed to be strictly synchronized, a synchronization processing flow is executed. First, for each acquisition timestamp recorded in the raw voltage and current data, the continuous readings of the three-phase voltage of the rolling mill drive motor and the continuous readings of the three-phase current of the rolling mill drive motor are matched one by one. The specific method is to set a timestamp matching tolerance, which is determined according to the maximum inherent time deviation of the data acquisition hardware and the accuracy requirements of the signal processing. For example, if the maximum physical delay or clock offset between the voltage and current sampling channels is estimated to be 5 microseconds, and considering the accuracy requirements of the subsequent phase calculation, the matching tolerance can be set The setting process is as follows: Set the maximum clock asynchrony between sampling channels of the system to 5 microseconds. To ensure the robustness of the match, a safety factor of If is 2, When the absolute value of the difference between the timestamps of two readings from different sensors is less than this 10 microsecond tolerance, they are considered to be paired data at the same sampling moment. Subsequently, the time base is unified, and the timestamp of one of the channels (for example, the timestamp of the voltage of phase A) or the average of all the timestamps involved in the matching is selected as the unified time base point of the synchronization moment, and data alignment is performed. If the corresponding readings of all six signals cannot be found within the set tolerance window, for example, the current data of a certain phase is missing, it is processed according to the preset missing data processing strategy, such as marking the data at that time point as incomplete or using linear interpolation to fill in the data based on the nearest valid synchronization data points under permitted conditions (for example, when a single sampling point is missing and the change is slow). The condition for linear interpolation is that the number of missing points does not exceed one, and the time interval between the previous and next data points does not exceed twice. ,After completing the above operations, all the voltage and current data that are ,successfully matched or effectively filled under a unified time reference are ,organized to form a one-to-one correspondence, and ,finally generate synchronized voltage and current data with a unified ,time axis.

[0053] Based on the synchronized voltage and current data generated in the previous step, although the data has been aligned in time, the measured amplitude and phase of each sensor may still have inherent or systematic errors introduced by environmental changes, so calibration is required. First, call the standard reference voltage signal and standard reference current signal obtained and stored during the equipment installation and commissioning phase or regular calibration and maintenance. These standard reference signals are obtained by connecting the rolling mill drive motor to a standard load with known characteristics (such as a pure resistive load) or by injecting a test signal with known amplitude and phase, and using a metrology-grade reference instrument (such as a power analyzer) to measure the voltage and current values. These values ​​are recorded as benchmarks, for example, the effective value of the standard reference voltage 400V, standard reference current effective value The output voltage of the sensor is 100A under specific test conditions. is 398V, current is 101A, phase difference The amplitude is 0.5 degrees (while it should be 0 degrees under standard load). Next, the amplitude comparison and phase difference compensation are performed on each sampling point data of each phase in the synchronized voltage and current data. The amplitude calibration is achieved by calculating the amplitude calibration coefficient, such as the voltage amplitude calibration coefficient , current amplitude calibration coefficient , and then multiply the synchronized voltage data by , current data multiplied by , the phase difference compensation is to calculate the phase correction amount , and apply this correction amount to subsequent phase-dependent calculations, such as by vector rotation or by directly adding or subtracting the fixed offset when calculating the phase angle. The update condition of the amplitude and phase calibration coefficients can set a "calibration coefficient drift threshold". For example, if any calibration coefficient or If the change from its initial calibration value exceeds 0.2%, an alarm is triggered to prompt recalibration. The setting of this 0.2% threshold is based on the requirement that the overall measurement accuracy does not exceed 0.5%, and takes into account the cumulative effect of multiple error sources. Through the above processing, the systematic errors of the voltage amplitude and current amplitude, as well as the voltage phase and current phase of each channel are corrected, thereby obtaining a calibrated electrical measurement set.

[0054] The steps to obtain the instantaneous motor shaft power sequence are:

[0055] Based on the calibrated electrical measurement set, the three-phase instantaneous voltage and current values ​​of the rolling mill drive motor are extracted. The data of each phase are combined one by one according to the phase synchronization sampling time. The three sets of voltage and current pairs at each sampling time in the combination are recorded to generate a three-phase instantaneous voltage and current combination sequence.

[0056] According to the three-phase instantaneous voltage and current combination sequence, the total instantaneous input electric power at each sampling moment is calculated. The calculation formula is:

[0057] ;

[0058] in, For the The total instantaneous input electrical power at the sampling moment, in watts (W), For the Sampling time The instantaneous voltage value of the phase, in volts (V), For the Sampling time The instantaneous current value of the phase, in amperes (A), is the phase sequence number, with values ​​of 1, 2, and 3, indicating three phases;

[0059] Based on the total instantaneous input electric power at each sampling moment, the calibrated efficiency curve data of the rolling mill drive motor at different load rates is called, the corresponding sampled power value is looked up in the table to obtain the loss power value, and the loss power value is subtracted from the total instantaneous input electric power item by item to generate the instantaneous motor shaft power sequence.

[0060] Specifically, based on the calibration electrical measurement set, the set is the synchronized time series data of the three-phase (A, B, C phase) voltage and three-phase current after amplitude and phase calibration, specifically including the sampling time Calibrated instantaneous voltage value 、 、 and calibrate the instantaneous current value 、 、 First, the instantaneous voltage and current values ​​of each phase at each synchronous sampling moment are accurately extracted from the set. For example, at the sampling moment , extracted 、 , 、 ,as well as 、 Then, the voltage and current data of each phase are paired one by one strictly according to their corresponding synchronous sampling moments to form voltage and current pairs. The specific operation is as follows: for each sampling moment , the A-phase voltage and A-phase current at this moment form a pair , the B-phase voltage and the B-phase current form a pair , the C-phase voltage and C-phase current form a pair ,Then, these three sets of voltage and current pairs are recorded as a whole and associated with their common sampling time This recording process is continuous and covers all sampling moments in the calibration electrical measurement set. In this way, the calibration data originally organized by signal type is reconstructed into a data structure organized by time and containing complete three-phase electrical information at each moment, thereby generating an ordered three-phase instantaneous voltage and current combination sequence. Each element of the sequence represents a sampling moment and contains the pairing information of the instantaneous voltage and instantaneous current of the three phases at that moment, providing a direct data input basis for the subsequent instantaneous power calculation.

[0061] formula: It provides basic data for real-time monitoring of motor operating status and energy efficiency analysis. By summing the products of three-phase instantaneous voltage and instantaneous current, it can accurately reflect the actual power consumption in the case of three-phase imbalance or waveform distortion.

[0062] For the Sampling time The instantaneous voltage value of the phase, in volts (V), is obtained from the "three-phase instantaneous voltage and current combination sequence" generated in the previous step. Each sampling moment record contains three-phase voltage and current pairs. That is the first Phase (e.g., Corresponding to phase A, Corresponding to phase B, For example, for a motor with a rated line voltage of 380V, the instantaneous value of its phase voltage can be about arrive Fluctuations between ), at a certain sampling moment By consulting the “three-phase instantaneous voltage and current combination sequence”, the instantaneous voltage of phase A can be obtained. .

[0063] For the Sampling time The instantaneous current value of the phase, in amperes (A), is also obtained from the "three-phase instantaneous voltage and current combination sequence" generated in the previous step, corresponding to the first The sampling time record The instantaneous current value of the phase changes significantly with the change of motor load. For example, a motor with a rated current of 100A at a certain sampling moment By consulting the “three-phase instantaneous voltage and current combination sequence”, the instantaneous current of phase A can be obtained. .

[0064] It is the phase sequence number, a dimensionless integer with values ​​of 1, 2, and 3, representing phase A, phase B, and phase C (or phase R, S, and T) of three-phase alternating current, respectively.

[0065] It is the index of the sampling moment, a dimensionless integer or timestamp, representing a specific discrete time point in the time series of data acquisition, which corresponds to the unique time identifier of each record in the "three-phase instantaneous voltage and current combined sequence".

[0066] Calculation process:

[0067] According to the three-phase instantaneous voltage and current combination sequence, at the sampling time , extract the instantaneous voltage and instantaneous current values ​​of each phase as follows:

[0068] For Phase A ( ): Obtained from the "three-phase instantaneous voltage and current combination sequence" and .

[0069] For phase B ( ): Obtained from the "three-phase instantaneous voltage and current combination sequence" and .

[0070] For Phase C ( ): Obtained from the "three-phase instantaneous voltage and current combination sequence" and .

[0071] Calculate the instantaneous power of each phase:

[0072] Phase A instantaneous power .

[0073] Phase B instantaneous power .

[0074] Phase C instantaneous power .

[0075] Calculate the total instantaneous input electrical power:

[0076] ;

[0077] ;

[0078] ;

[0079] This result shows that at the sampling time , the total instantaneous input power of the rolling mill drive motor is Watt (or kilowatts), this calculated The value is a data point in the "instantaneous electric power data" sequence, and the entire sequence consists of This calculation process is repeated and the sequence is the basis for subsequent estimation of motor losses and instantaneous motor shaft power.

[0080] The total instantaneous input electric power at each sampling moment calculated based on the previous step , namely "instantaneous electric power data", then call the calibrated efficiency curve data of the rolling mill drive motor at different load rates. The calibrated efficiency curve data is usually provided by the motor manufacturer, or obtained and digitally stored through standard load tests before the motor leaves the factory and during regular maintenance (for example, using a dynamometer to accurately measure the output torque and speed at different input powers, so as to calculate the output power and motor loss). Its content is generally a series of corresponding relationships, describing the specific values ​​of various losses (including copper loss, iron loss, mechanical loss, etc.) or total losses of the motor at different input electric powers or equivalent load percentages. For example, a data table may contain the following columns: input power (kW), total loss power (kW). When querying, the current instantaneous input electric power As a reference for table lookup, find the corresponding power loss value in the calibration efficiency curve data. If the value of exactly matches a certain input power point in the datasheet, the power loss value of that point is directly used. The value of is between two adjacent input power points in the data table, for example Corresponding loss , Corresponding loss ,and , the linear interpolation method is used to estimate Corresponding power loss value , calculated as ,if If the power exceeds the coverage of the calibration data table, for example, it is lower than the minimum recorded power or higher than the maximum recorded power, it will be processed according to the preset boundary processing rules. The rules are set based on the physical characteristics of the motor. For example, for the case below the minimum recorded power, the no-load loss value or the loss value of the minimum recorded point can be used. For the case above the maximum recorded power, the trend can be extrapolated based on the last few data points, but at the same time, it is marked that the estimated loss of the data point may have a large uncertainty, and the loss at each sampling moment is obtained. Corresponding estimated motor losses Then, the corresponding total instantaneous input power is Deduct the motor loss from the calculation to calculate the instantaneous motor shaft power ,The data at each sampling moment is processed in this way, and finally a time series is generated, namely the instantaneous motor shaft power series.

[0081] The steps to obtain the key harmonics of the shaft power spectrum are:

[0082] Based on the instantaneous motor shaft power sequence, continuous power value segments in the sequence are intercepted at fixed time intervals, and the time starting point and ending point of each segment are recorded to form a periodically segmented instantaneous shaft power segment sequence;

[0083] According to the instantaneous shaft power segment sequence divided into periodic segments, Fourier transform is performed on each shaft power segment sequence one by one, the shaft power value is mapped from the time domain to the frequency domain, and the frequency domain power spectrum data of each shaft power segment sequence is generated;

[0084] Based on the frequency domain power spectrum data, the defined target frequency points are retrieved, the amplitude value and phase value corresponding to each target frequency point are extracted from the frequency domain power spectrum data, and the key harmonics of the axis power spectrum are established.

[0085] Specifically, based on the instantaneous motor shaft power sequence generated in the previous step, this sequence represents the instantaneous power value of the actual output shaft of the rolling mill drive motor. Over time A set of continuously changing discrete data points with a specific sampling frequency. For example, the sampling frequency of this sequence is set to First, in order to carry out subsequent frequency domain analysis, it is necessary to use a "fixed time interval" Continuously extract segments containing continuous power values ​​from the instantaneous motor shaft power sequence. The "fixed time interval" The selection of is an important parameter, and its setting needs to balance two aspects. On the one hand, It needs to be long enough to include several complete cycles of the lowest fault characteristic frequency to be analyzed (such as the rotation frequency of a low-speed shaft of the reducer or its harmonics), so as to ensure the frequency resolution of the subsequent Fourier transform ( ) is high enough to distinguish close frequency components, for example, if the lowest frequency of interest , and it is expected to capture at least 10 cycles to obtain a stable spectrum estimate, then it is required Seconds, on the other hand, It should not be too long to ensure the quasi-stationary nature of the signal within the segment. For example, the statistical characteristics of the signal do not change dramatically within this period of time. In actual operation, the number of data points is often combined with the subsequent fast Fourier transform (FFT) algorithm. preference (usually a power of 2 for computational efficiency), to determine , if the number of data points in the selected segment is The actual length of the segment 2.048 seconds is the selected "fixed time interval". When intercepting, you can use the data window overlap method. For example, set the overlap rate to 50%, which means that every A new power value segment of 2.048 seconds in length is intercepted from the instantaneous motor shaft power sequence. For each continuous power value segment intercepted, the timestamp or index number corresponding to the starting sampling point in the original time series and the timestamp or index number corresponding to the ending sampling point are accurately recorded. Through the above segmentation and recording process, the original continuous instantaneous motor shaft power sequence is converted into a series of instantaneous shaft power segment sequences with clear start and end time marks and possibly overlapping, providing a standardized data input unit for subsequent spectrum analysis.

[0086] According to the periodic segmented instantaneous shaft power fragment sequence obtained in the previous step, each fragment is a segment of length (For example The discrete time series of the instantaneous shaft power value of the motor within a specific 2.048 seconds is represented by a number of points. Next, it is necessary to perform Fourier transform processing on these shaft power fragments one by one to map these power values ​​from the time domain to the frequency domain. This process is usually implemented using the fast Fourier transform (FFT) algorithm with high computational efficiency. Before performing FFT on each fragment, in order to reduce the spectral leakage effect caused by signal truncation (that is, the phenomenon of energy spreading from the true frequency to the adjacent frequency), a window function is usually applied to the fragment data. For example, the Hanning window is selected. The Hanning window is widely used because of its good compromise performance in terms of main lobe width and sidelobe suppression. Its selection is based on the consideration of the expected signal characteristics (such as whether there are strong interfering frequencies) and the analysis objectives (such as whether extremely high frequency resolution or extremely large dynamic range is required). The window function will be multiplied point by point with each power value in the fragment to smoothly attenuate the data at both ends of the fragment. The processed fragment data is then input into the FFT algorithm. The output of the FFT algorithm is a series of complex numbers. The number of these complex numbers is equal to the number of points in the input fragment. Similarly, each complex number corresponds to a specific discrete frequency point (or frequency bin), which ranges from 0Hz (DC component) to Nyquist frequency (ie sampling frequency). half, for example ), the interval between frequency points, that is, the frequency resolution , by the clip duration Decisions, e.g. After completing the FFT calculation of a shaft power segment sequence, the result is the frequency domain power spectrum data of the segment, which contains the amplitude information and phase information of the shaft power signal at each discrete frequency point (usually stored in complex form, with the real and imaginary parts jointly determining the amplitude and phase).

[0087] The frequency domain power spectrum data generated after Fourier transform is performed on each axis power segment sequence in the previous step is a data containing A sequence of complex values, each of which corresponds to a specific frequency bin and its amplitude and phase information at that frequency bin. Next, it is necessary to retrieve and extract the spectrum information related to the pre-defined "target frequency points" from this complete frequency domain power spectrum data. These "target frequency points" are a list that is pre-set and stored through theoretical calculation or empirical summary based on the specific mechanical structure parameters of the rolling mill reducer (for example, the number of teeth of each gear stage, the precise speed of each shaft), motor parameters (such as pole pairs, power supply frequency) and known characteristic frequencies related to common failure modes (such as gear wear, broken teeth, pitting of bearing inner and outer rings, rolling element damage, cage fracture, etc.). For example, if the motor speed is 900 rpm, its fundamental frequency is , which may be a target frequency point. If the number of teeth on the driving gear of a certain gear is , then its meshing frequency It will also be listed as a target frequency point. For each target frequency point in the preset list , it is necessary to find the actual frequency bin closest to it in the discrete frequency domain power spectrum data of the current segment (in , is the frequency bin index, is the frequency resolution), since May not be exactly equal to any , so a matching tolerance is usually used to determine the corresponding relationship. This tolerance is generally set to half of the frequency resolution, that is, , the closest one that satisfies this condition The corresponding frequency bin index is selected, then, from that selected frequency bin The corresponding complex spectrum value , calculate and extract the amplitude values ​​respectively (It may be necessary to correct the amplitude and phase values ​​according to the spectrum type, such as single-sided spectrum or double-sided spectrum, and the influence of the window function to obtain the engineering significance) , organize all target frequency points and their corresponding extracted amplitude and phase values ​​to form a structured list, such as , this list is the key harmonic data of the shaft power spectrum established for the current shaft power segment.

[0088] The steps to obtain the current in the frequency band of interest are:

[0089] The amplitude of each sampling point of the stator current signal of the rolling mill drive motor is read from the calibration electrical measurement set and arranged in chronological order. The current amplitude of all sampling points in the last 30 seconds is intercepted to form a sampling segment. The maximum amplitude, average amplitude, total number of sampling points and fixed sampling interval value in the signal segment are extracted to generate a time-limited stator current amplitude sequence.

[0090] Based on the stator current amplitude sequence in the time-limited period, the filter center frequency judgment value is calculated using the following formula:

[0091] ;

[0092] in, Indicates the filter center frequency judgment value, the unit is Hertz (Hz), is the stator current sampling time interval, in seconds (s), For the The stator current amplitude at each sampling point, in amperes (A), is the average value of the stator current amplitude sequence in the time limit, in amperes (A), is the maximum amplitude in the sequence, in amperes (A). is the total number of sampling points in the sequence, is the empirical correction factor used to correct the center frequency calculation;

[0093] Based on the filter center frequency judgment value, the upper and lower cutoff frequencies of the bandpass filter are set with the filter center frequency judgment value as the center, the cutoff range is set using the fixed bandwidth mode, and the filter is executed to perform bandpass processing on the spectrum of the stator current amplitude sequence in the time period to form the frequency band of interest.

[0094] Specifically, from the calibrated electrical measurement set obtained in the previous step, which is a time series data containing the synchronized and amplitude- and phase-calibrated instantaneous voltage and instantaneous current of the three phases (e.g., phases A, B, and C) of the rolling mill drive motor, we first focus on the stator current signal of the rolling mill drive motor and select the calibrated instantaneous current value of one phase (e.g., phase A). As a representative stator current signal, or using a certain combined amplitude of the three-phase current (such as the instantaneous value of the current vector modulus), here for clarity, the absolute value of the instantaneous current of phase A is used. As the sampling point amplitude , these current amplitudes Arrange them in sequence according to their inherent time sequence to form a continuous time series. Then, extract the data of the most recent fixed time period from this complete time series. The time period is set to 30 seconds based on experience. The basis for selecting 30 seconds is that this time period is sufficient to include multiple complete power frequency cycles and potential low-frequency modulation phenomena of the motor under various working conditions (including smooth operation, load fluctuations, etc.). At the same time, it is relatively short and can reflect the recent motor operating status and avoid long-term averaging to cover up sudden characteristics. For all current amplitude sampling points within this intercepted 30-second time period, a sampling segment is formed, and then the statistical characteristics of the data in the sampling segment are extracted. The specific extracted parameters include: the maximum current amplitude in this sampling segment , calculated by traversing all sampling points in the segment and find the maximum value, e.g. ; Calculate the arithmetic mean of all sampling points in the segment to get the average current amplitude ,For example ; Count the total number of sampling points in this section , if the sampling frequency of the original calibration electrical measurement set is , then the number of sampling points contained in the 30-second time period is At the same time, record the fixed sampling time interval used by the sampling segment , whose value is the inverse of the sampling frequency, that is, , these extracted current amplitude sequences themselves and their statistical parameters 、 、 and Together they constitute a time-limited stator current amplitude sequence, providing input for the subsequent calculation of the filter center frequency determination value.

[0095] formula: The benefit of the formula is that it provides an adaptive method based on the time-domain statistical characteristics of the current signal to estimate a characteristic frequency, which can be used to guide the subsequent band-pass filter design to extract specific frequency band information related to the device status. The formula comprehensively considers the fluctuation degree of the current amplitude ( ), the relative magnitude of the current amplitude ( ) and the overall energy distribution characteristics of the signal, try to locate the equivalent frequency corresponding to the more significant dynamic changes in the signal, and the empirical correction factor The introduction of increases the robustness of the formula and its adaptability to specific application scenarios.

[0096] The stator current sampling interval is in seconds (s). This parameter is determined in the previous step "Generate a time-limited stator current amplitude sequence" and comes directly from the sampling settings of the calibration electrical measurement set. In the previous example, if the sampling frequency ,but , which represents the discretization accuracy of the current signal on the time axis.

[0097] is the first in the “time-limited stator current amplitude sequence” The stator current amplitude (here refers to the modulus of the current, ensure it is non-negative) of the sampling points is in amperes (A). These values ​​are the actual data points that constitute the sequence generated in the previous step, reflecting the instantaneous magnitude of the motor stator current in the last 30 seconds. For example, the sequence contains .

[0098] is the arithmetic mean of the “limited time period stator current amplitude sequence” in amperes (A). This value has been calculated in the previous step and represents the average level of the current amplitude within the limited time period. For example, based on the above data, we can calculate .

[0099] is the maximum value in the “limited time stator current amplitude sequence”, in amperes (A). This value has also been extracted in the previous step and represents the peak level of the current amplitude within the limited time period. For example, .

[0100] is the total number of sampling points included in the “time-limited stator current amplitude sequence”, which is a dimensionless integer and has been determined in the previous step. For example, for a 30-second time period and a 20 kHz sampling frequency, .

[0101] It is an empirical correction factor used to correct the center frequency calculation. It is a dimensionless constant. Its specific value is set based on a large amount of historical data analysis and experimental verification of the current signal characteristics of a specific type of rolling mill drive motor under various healthy and fault conditions. The selection principle is to make the calculated It can most effectively point to the characteristic frequency bands related to typical faults (such as current modulation caused by bearing faults and gear faults), or provide a stable reference frequency band when there are no obvious fault features. For example, by retrospectively analyzing 100 sets of motor current data containing known faults, the frequency bands of different Values ​​(such as multiple values ​​ranging from 0.0001 to 0.1) The consistency with the actual fault characteristic frequency and the The stability of As an optimized value for this type of device, this value helps avoid undersized or unstable calculation results when signal fluctuations are small, and helps fine-tune the results.

[0102] Calculation process: Due to the actual number of sampling points It is very large (for example, 600,000), and it is impossible to list the calculations item by item. Here, a small data set is used for schematic calculations. The parameter values ​​are as follows: , (corresponding to sampling frequency 1000Hz). Take from the “limited time stator current amplitude sequence” , , .but . . Experience correction factor .

[0103] Calculate each element inside the sum: : ;

[0104] ;

[0105] for : ;

[0106] ;

[0107] for : ;

[0108] ;

[0109] Sum:

[0110] ;

[0111] Calculate the average:

[0112] ;

[0113] Add the empirical correction factor and take the square root:

[0114] ;

[0115] Multiply :

[0116] ;

[0117] The results show that according to the input "limited time stator current amplitude sequence" and its statistical characteristics, combined with the empirical correction factor, the calculated filter center frequency determination value is This value will be used as the basis for setting the center frequency of the bandpass filter in the future to extract the specific frequency band information around this frequency in the signal.

[0118] Based on the filter center frequency determination value calculated in the previous step (For example, by calculating ), and the time-limited stator current amplitude sequence generated in the first step (i.e., the current amplitude with a duration of 30 seconds) The time series of the bandpass filter is then designed and applied. First, the filter center frequency is used to determine the value As the theoretical center frequency of the bandpass filter Then, the fixed bandwidth mode is used to set the passband range of the filter. It is a preset parameter, which is determined based on the prior knowledge of the current modulation signal characteristics generated by common faults of the rolling mill drive motor and its reducer (such as bearing damage, gear wear, etc.). These fault characteristics are often manifested as a certain carrier frequency (which may be related to the motor rotation frequency, meshing frequency, etc., or may be related to The estimated characteristic frequency band is related to the sideband components within a certain range. By analyzing the current spectra of a large number of fault cases, a bandwidth is selected that can effectively cover these typical fault characteristic sidebands while being too wide to avoid introducing too much noise. For example, a fixed bandwidth is set. The selection of this value is based on historical data analysis. It is found that under various fault conditions, the key modulation information is usually distributed within the range of 25Hz above and below the center frequency. and bandwidth Then, calculate the lower cutoff frequency of the bandpass filter and upper cutoff frequency ,Right now ,as well as , if the calculated If it is less than or equal to zero, it is set to a very small positive frequency value close to zero (for example, 1 Hz) to ensure physical meaning. Then, a specific type of digital filter is selected, such as an 8th-order Butterworth bandpass filter, because it has the flattest amplitude-frequency response characteristics in the passband and good roll-off characteristics. Then, the designed bandpass filter is applied to the time-limited stator current amplitude sequence (time domain signal). Usually, the time domain sequence is first subjected to a fast Fourier transform (FFT) to obtain its spectrum, and then the spectrum is multiplied by the frequency response function of the designed bandpass filter in the frequency domain (that is, in arrive The value within the frequency range is 1 or close to 1, and the value in the rest of the frequency range is 0 or close to 0, and it decays smoothly in the transition band). Finally, an inverse fast Fourier transform (IFFT) is performed on the multiplied spectrum to obtain the filtered time domain signal. This time domain current signal after bandpass filtering is the current in the frequency band of interest.

[0119] The steps to obtain current impact and cycle characteristics are:

[0120] Extract all sampling point data from the current in the frequency band of interest, perform Hilbert transform point by point, and modulo the results to obtain the current envelope amplitude corresponding to each sampling point, forming a frequency-selective current envelope sequence. At the same time, extract the maximum envelope amplitude, average amplitude, and original amplitude of each sampling point to form a normalized envelope amplitude structure;

[0121] Based on the normalized envelope amplitude structure, the spectrum kurtosis value is calculated using the following formula:

[0122] ;

[0123] in, is the spectral kurtosis value (dimensionless), Indicates the first frequency in the frequency-selective current envelope sequence The amplitude of each sampling point, in amperes (A), is the average amplitude of the sequence, in amperes (A), is the maximum amplitude in the sequence, in amperes (A). is the total number of sampling points, is the kurtosis correction factor;

[0124] Based on the spectrum kurtosis value, it is determined whether it is higher than the periodic shock recognition threshold. If it is higher than the periodic shock recognition threshold, it is determined that there is a high-amplitude shock feature. Then, combined with the uniformity of the time interval between consecutive local peaks, the periodic repeatability is judged, and the current shock and periodic characteristics are output.

[0125] Specifically, the frequency band current obtained in the previous step is the time domain signal sequence of the stator current of the rolling mill drive motor after being processed by a specific bandpass filter. First, all the discrete sampling point data are extracted from this frequency band current, which is recorded as ,in is the sampling point index. Then, in order to obtain the envelope information of the current in the frequency band of interest, Perform Hilbert transform on the sequence point by point to obtain its Hilbert transform sequence , then construct the analytical signal ,in is an imaginary unit, and then each sampling point of the analytical signal is modulo (calculated its amplitude), the specific calculation method is ,in is the sampling point index of the envelope sequence (with Corresponding to), the real number sequence obtained thereby This is the frequency-selective current envelope sequence, which reflects the slow change characteristics of the current signal amplitude in the frequency band of interest. (Its length is sampling points, which is the same as the length of the input frequency band current segment) to perform statistical parameter extraction, including calculating and recording the maximum amplitude in the envelope sequence (For example, by iterating over all get ), calculate and record the arithmetic mean amplitude of the envelope sequence (For example, ), and the maximum amplitude of these extracted envelopes , average amplitude , and the envelope sequence itself (i.e. the original envelope amplitude of each sampling point) and the total number of sampling points They are organized together to form a structured data set, called the normalized envelope amplitude structure, which will serve as the basic data for the subsequent calculation of the spectral kurtosis value.

[0126] formula: The benefit of this formula is that the adjusted kurtosis value calculated by this formula is It can effectively characterize the intensity of the impact component in the frequency-selective current envelope sequence. Compared with the standard kurtosis, this formula introduces the normalization process based on the maximum value and the weighting of the amplitude itself ( ), which aims to enhance the sensitivity to impact signals occurring at higher energy levels while maintaining the ability to measure the deviation of the overall distribution of the signal from the Gaussian distribution. This characteristic makes it potential in detecting weak impact signals generated by early local damage of components such as rolling bearings and gears. The kurtosis correction factor The introduction of provides flexibility for calibration and optimization according to specific equipment and signal characteristics in practical applications. Its design idea is to correct and normalize the ratio of the fourth-order central moment of the envelope signal to the square of the second-order central moment to highlight the "peak" level of the signal.

[0127] Indicates the first frequency in the frequency-selective current envelope sequence The amplitude of each sampling point is in amperes (A). These data points come from the "frequency-selective current envelope sequence" generated in the previous step. They are the instantaneous values ​​of the current envelope obtained after Hilbert transform and modulo operation. For example, in an envelope sequence containing an impulse signal, the following may be observed: , (impact point), Equal values.

[0128] is the arithmetic mean amplitude of the "frequency-selected current envelope sequence", in amperes (A). This value has been calculated in the previous step "Forming the normalized envelope amplitude structure" and represents the average intensity of the envelope signal during the observation period. For example, if a section of envelope data mainly fluctuates between 0.1A and 0.5A, but a few impulses reach 1.5A-2.0A, then its average value may be .

[0129] The maximum amplitude in the "frequency-selective current envelope sequence" is in amperes (A). This value has also been extracted in the previous step and represents the peak level that the envelope signal can reach during the observation period. For signals containing impulses, this value is usually much larger than the average value, for example .

[0130] is the total number of sampling points in the "frequency-selective current envelope sequence", which is a dimensionless integer. Its value is the same as the number of sampling points of the "interest band current" segment in the previous step. For example, if the "interest band current" comes from a signal segment with a duration of 30 seconds and a sampling frequency of 20 kHz, then point.

[0131] is the kurtosis correction factor, which is a dimensionless constant. The setting of this factor is based on a large number of known equipment states (healthy and various faults). The calculation results are statistically analyzed and optimized to adjust the calculated kurtosis value to a more discriminative range, or to make it comparable with a certain reference standard (such as the kurtosis of Gaussian distribution is 3). For example, the current envelope data of 100 groups of healthy devices and 100 groups of specific early fault devices are collected to calculate their uncorrupted current envelope values. The corrected kurtosis value is found to be distributed in the healthy equipment Range, faulty equipment is distributed in Range, if you want to adjust the baseline of the health state to a specific value (e.g. ), you can set , or set , the original calculation result in the brackets is directly used as the kurtosis value, and the subsequent discrimination threshold is set entirely based on the statistical distribution of the original calculation result. Here, .

[0132] Calculation process:

[0133] Here we use a small dataset ( ) is used for schematic calculation, and the parameter values ​​are as follows:

[0134] Get from the "Normalized Envelope Amplitude Structure":

[0135] , , , .

[0136] .

[0137] .

[0138] Total number of sampling points .

[0139] Kurtosis correction factor .

[0140] Calculate the summation term in the numerator ,in :

[0141] ,but .

[0142] ,but .

[0143] ,but .

[0144] ,but .

[0145] Numerator summation .

[0146] Calculates the square of the sum in the denominator ,in :

[0147] ,but .

[0148] ,but .

[0149] ,but .

[0150] ,but .

[0151] Sum within the denominator .

[0152] Denominator .

[0153] calculate :

[0154] ;

[0155] The results show that for this example signal segment, the adjusted and corrected spectral kurtosis value is This value reflects the impact characteristics of the signal envelope and will be compared with the preset "periodic impact identification threshold" to determine whether there is a high-amplitude impact.

[0156] Based on the adjusted spectral kurtosis value calculated in the previous step (For example, ), first compare it with a preset "periodic shock identification threshold" For comparison, the threshold It is obtained based on the statistical analysis of historical current envelope data of a large number of rolling mill reducers in different health states and fault modes. The specific setting process is: collect at least two types of sample data, one type is when the equipment is operating normally The other type is when the equipment has a known periodic impact fault (such as pitting of the bearing outer ring and broken gear teeth). value sets, analyze the statistical distribution of the two sets (such as mean, variance, range), and select a boundary that can effectively distinguish the two as For example, if the normal device Usually less than 4.0, while the value of the equipment with impact failure is generally greater than 7.0, so it can be initially set , and further optimize the threshold through ROC curve analysis and other methods to balance the detection accuracy and false alarm rate. (such as 0.5159) is lower than (As in 5.5), it is preliminarily determined that there is no significant high-amplitude impact feature. If Higher than , it is determined that there is a high-amplitude impact feature. In the case of determining that there is a high-amplitude impact feature, the "frequency-selective current envelope sequence" is further Analysis is performed to determine the periodic repeatability of the shock. This process involves first using a peak detection algorithm (for example, finding a local maximum point in the sequence that is larger than its left and right neighbors and exceeds a certain dynamic threshold) to identify all significant local peak points in the envelope sequence, recording the time (or sampling point index) of these peak points, and then calculating the time interval sequence between consecutive (or specific amplitude conditions) local peaks. , then, by calculating the statistical parameters of these time intervals, such as the average interval and standard deviation , to assess its uniformity, for example, to calculate the coefficient of variation and compare it with a preset "uniformity judgment threshold" (For example This value is based on the tolerance setting of the impact interval jitter in the typical periodic fault signal. If the interval jitter of the real periodic impact generally does not exceed 10% of the average interval, then 0.15 can be used as a looser judgment basis for comparison. , then the impact is considered to have strong periodic repeatability. Finally, the judgment results of the high-amplitude impact characteristics and the periodic repeatability are combined to output a comprehensive description of the current impact and periodic characteristics of the current signal segment, such as {impact intensity: low, periodicity: not applicable} or {impact intensity: high, periodicity: strong, average impact period: Second}.

[0157] The steps to obtain the multi-dimensional electrical state vector are:

[0158] Call the key harmonics of the shaft power spectrum, extract the harmonic amplitude value and harmonic phase value corresponding to each target frequency point in the key harmonics of the shaft power spectrum, and pair them one by one in the order of frequency points to generate shaft power spectrum feature data containing amplitude and phase;

[0159] Based on the shaft power spectrum characteristic data, the current impact and period characteristics are called, the spectrum kurtosis value and cycle frequency characteristic data in the current impact and period characteristics are extracted one by one, and time synchronization is performed with the shaft power spectrum characteristic data according to the corresponding time to generate synchronized electrical characteristic data;

[0160] Based on the synchronized electrical characteristic data, the harmonic amplitude, harmonic phase, spectral kurtosis value and cyclic frequency characteristics are arranged in order, and a characteristic parameter vector with time series correlation is constructed by combining them one by one to obtain a multidimensional electrical state vector.

[0161] Specifically, the key harmonic data of the shaft power spectrum established in the previous step is called, which is for each analysis time period (for example, the first harmonic of the instantaneous motor shaft power sequence). A set of target frequency points and their corresponding amplitudes and phases is extracted from the first segment. For example, The specific form of the key harmonic data of the axis power spectrum is ,in For the total number of preset target frequency points, first, for each record in the key harmonic data of the axis power spectrum, that is, for each target frequency point, extract the corresponding harmonic amplitude value (for example , , etc.) and harmonic phase values ​​(e.g. , , etc.), then, these extracted harmonic amplitudes and corresponding harmonic phases will be paired one by one strictly according to the predefined order of "target frequency points". This order is fixed, for example, they are arranged from low to high according to the target frequency points, or according to their specific monitoring importance order defined in the system configuration, ensuring the consistency of the elements in the subsequent feature vectors. Through this pairing operation, for the first An ordered list is formed, in which each element is a value pair of (harmonic amplitude, harmonic phase), and the length of the list is equal to the total number of target frequency points. , the generated ordered list is the shaft power spectrum characteristic data containing amplitude and phase. This data structure clearly describes the energy distribution and phase relationship of the motor shaft power at each key frequency during the time period.

[0162] Based on the previous step, for each analysis time period (e.g. The axis power spectrum characteristic data containing amplitude and phase prepared for the analysis time period) is used, and the current impact and period characteristics obtained by analyzing the current signal for the same time period in the previous step are called. The current impact and period characteristics are structured data containing multiple diagnostic indicators, for example, it contains the calculated spectrum kurtosis value. , and the determination of whether a shock exists ( ), whether the shock is periodic ( ), if it is periodic, it also includes its average impact period (that is, the cyclic frequency characteristic data here If there is no periodic shock, this value can be set to a predefined specific value, such as 0 or an invalid identifier). First, the required specific values, namely the spectrum kurtosis value, are extracted one by one from the current shock and periodic feature data structure. and cycle frequency characteristic data Subsequently, these features (spectral kurtosis value and cycle frequency characteristic data) obtained from the current analysis path are time-synchronized with the shaft power spectrum characteristic data obtained from the power analysis path. The basis of time synchronization is to ensure that all these features correspond to the same raw data acquisition and analysis time window of the rolling mill operation (for example, Analysis time period), because at the system design level, the feature extraction process for power signals and current signals is based on data segments with aligned or identical start and end times. Therefore, in this step, the feature data from these two sources that describe the device status in the same time period are mainly aggregated and associated to ensure that all elements in the subsequently constructed multidimensional vector accurately reflect the device operating status at the same time. After completing the above extraction and synchronous association, the analysis time period is generated. The synchronized electrical characteristic data is a more comprehensive feature set that integrates power spectrum information and current envelope impact characteristics.

[0163] Based on the previous step, for each analysis time period (e.g. Time period) integrated synchronized electrical characteristic data, which includes the axis power spectrum characteristic data in the time period (an ordered list of P pairs of harmonic amplitudes and harmonic phases, where P is the number of target frequency points), spectrum kurtosis value and cyclic frequency characteristic data, and then arrange these characteristic elements in a predetermined specific order, which defines the structure of the final feature vector. For example, the order can be set as: first the harmonic amplitude of the first target frequency point, then its harmonic phase, then the harmonic amplitude of the second target frequency point, then its harmonic phase, and so on, until all The amplitude and phase of each target frequency point are arranged in sequence, and then, following these power harmonic characteristics, the spectrum kurtosis value is arranged, and finally the cyclic frequency characteristic data is arranged. According to this strict order, each specific value in the synchronized electrical characteristic data is extracted and combined one by one, thus providing the current analysis time period. Construct a feature parameter vector with fixed dimension and fixed element order , for example, if there is target frequencies, then the vector Since each of these characteristic parameter vectors corresponds to a specific analysis period that is continuous in time or advances in a fixed step, these successively generated characteristic parameter vectors Arranging them in their corresponding time sequence forms a characteristic parameter vector sequence with time series association. Each vector in this sequence is a multi-dimensional electrical state vector, which quantifies the operating status of the rolling mill reducer at the corresponding moment from multiple angles.

[0164] The steps for determining the working condition of the rolling mill reducer are as follows:

[0165] Extracting the value of each characteristic parameter in the multidimensional electrical state vector, respectively calling the pre-set reference threshold of the corresponding characteristic parameter, calculating the numerical difference between each characteristic parameter and the corresponding reference threshold one by one, and generating a numerical difference sequence between the multidimensional electrical state vector and the reference threshold;

[0166] Based on the difference value sequence between the multi-dimensional electrical state vector and the reference threshold, determine whether the absolute value of each difference value exceeds the set fluctuation amplitude threshold one by one, record the number and category of characteristic parameters that exceed the fluctuation amplitude threshold, and generate statistical results of characteristic parameters with excessive fluctuation amplitude;

[0167] Based on the statistical results of the characteristic parameters of the fluctuation amplitude exceeding the limit, the number and type of the characteristic parameters exceeding the limit in the statistical results are analyzed. If the number of characteristic parameters exceeding the preset warning threshold and the type includes periodic characteristics or harmonic characteristics, the operating condition of the rolling mill reducer is judged to be abnormal; otherwise, the operating condition is judged to be normal, thus forming a rolling mill reducer operating condition judgment.

[0168] Specifically, the previous step is extracted for each analysis time period (e.g. time periods) to construct a multidimensional electrical state vector The current value of each characteristic parameter in the vector For example, the elements contained are ,in is the number of target frequency points in the preset key harmonics of the shaft power spectrum, and Respectively The harmonic amplitude and phase of each target frequency point, is the spectral kurtosis value of the time period, is the cyclic frequency characteristic data of the time period, then, for each characteristic parameter in this multi-dimensional electrical state vector (in is the index of the parameter in the vector, from 1 to ), respectively call a pre-set benchmark threshold corresponding to the specific feature parameter , these benchmark thresholds It is determined by statistically analyzing the historical multi-dimensional electrical state vector data collected from the rolling mill reducer under a large number of known healthy operating conditions. The specific setting process is: for the first characteristic parameters, and collect them in The numerical sequence of healthy samples, calculate the mean of the sequence and standard deviation , then the baseline threshold It is usually set to the mean value under the health state, that is, , for example, for the harmonic amplitude of the first target frequency , if its mean value in healthy state is Unit, then this is the baseline threshold of the feature, for the spectral kurtosis value , if its mean value in healthy state is , then this is the benchmark threshold of the kurtosis feature. After completing the call of the benchmark threshold, calculate the value of each characteristic parameter in the current multidimensional electrical state vector one by one The corresponding benchmark threshold The numerical difference between , all these calculated difference values ​​are arranged and combined according to the order of the parameters in the original multidimensional electrical state vector, thereby generating a difference value sequence between a multidimensional electrical state vector with the same dimension as the original vector and the reference threshold, which is recorded as .

[0169] The numerical sequence of differences between the multidimensional electrical state vector generated in the previous step and the reference threshold , where each element Representative The degree to which a characteristic parameter deviates from its healthy baseline, then, for each difference value in the difference value sequence, Make judgments one by one, specifically calculate their absolute values and compare it with a parameter for this specific feature Pre-set "fluctuation threshold" For comparison, the fluctuation amplitude threshold It defines the maximum fluctuation range allowed for the corresponding characteristic parameter under normal operating conditions. The setting method is usually based on the statistical distribution characteristics of the characteristic parameter in a healthy state. For example, it can be set as the standard deviation of the characteristic parameter in a healthy state. multiples of times), i.e. ,in is the sensitivity factor, and its value (e.g. or ) is determined based on the balance between the required monitoring sensitivity and the acceptable false alarm rate, for example, the harmonic amplitude of the first target frequency , if the standard deviation in health state Unit, and set , then its fluctuation amplitude threshold Unit, for spectral kurtosis value , if the health standard deviation , then its fluctuation amplitude threshold If the judgment result is , then it is considered that If the fluctuation of a characteristic parameter exceeds the limit, the relevant information of the exceeding characteristic parameter is recorded, including its specific category (for example, "harmonic amplitude", "harmonic phase", "spectral kurtosis value" or "cyclic frequency characteristic", which are determined according to the predetermined position and meaning of the characteristic parameter in the multidimensional electrical state vector) and its index At the same time, all the characteristic parameters that exceed the limit are counted to obtain the total number of characteristic parameters with fluctuation amplitude exceeding the limit in the current analysis time period. These records (the number of exceeding parameters and their respective category lists) together constitute the statistical results of the characteristic parameters with fluctuation amplitude exceeding the limit in this time period.

[0170] Based on the statistical results of the characteristic parameters of fluctuation range exceeding the limit generated in the previous step, the result includes the total number of characteristic parameters with fluctuation range exceeding the limit in the current analysis period. and a list of the categories of each of these out-of-limit parameters (e.g., [ , ]), the following is an in-depth analysis of this statistical result. First, the number of over-limit characteristic parameters With a "preset warning threshold" For comparison, the It is an integer representing the minimum number of out-of-limit features that triggers the judgment of abnormal equipment working condition. Its setting is mainly based on expert experience and analysis of historical failure cases. It aims to distinguish occasional, unimportant parameter fluctuations from multi-parameter coordinated anomalies that indicate real failures. For example, through analysis of historical data, it is found that when at least three key categories of characteristic parameters exceed the limit at the same time, the probability of significant equipment failure is high. Therefore, it can be set ,like Not exceeded (For example, the current ,and ,but does not exceed the threshold), then it is directly determined that the rolling mill reducer is operating normally. Exceeded , then it is necessary to further check the type of the out-of-limit parameter, specifically to determine whether the category list contains at least one characteristic type that is considered to be a key indicator, namely "periodic characteristics" (here corresponding to "cyclic frequency characteristic data") or "harmonic characteristics" (here corresponding to "harmonic amplitude" or "harmonic phase"). If And the type of over-limit parameters does include at least one of the above key indicative characteristic types, then the comprehensive judgment of the rolling mill reducer working condition is abnormal. On the contrary, if However, the types of all out-of-limit parameters do not belong to the predefined key indicative types (this situation is rare in the feature structure of this application because the main features are all key types), or Not exceeded , it is determined that the working condition of the rolling mill reducer is normal. Finally, according to the above judgment logic, the working condition judgment result of the rolling mill reducer in the current analysis time period is output ("normal working condition" or "abnormal working condition").

Claims

1. A rolling mill reducer monitoring and diagnosis system based on multi-source information fusion, characterized in that: The system comprises: The electrical parameter signal acquisition module obtains continuous readings of the three-phase voltage and three-phase current of the rolling mill drive motor, synchronizes the readings, and performs amplitude and phase calibration to obtain a calibrated electrical measurement set; a shaft power characteristic calculation module, calling the three-phase voltage and the three-phase current of the rolling mill drive motor in the calibration electrical measurement set, calculating the instantaneous input electric power, obtaining instantaneous electric power data, estimating the motor loss, and deducting the motor loss from the instantaneous electric power data to obtain an instantaneous motor shaft power sequence, performing a Fourier transform on the instantaneous motor shaft power sequence, and establishing key harmonics of the shaft power spectrum; a current modulation analysis module, which extracts a rolling mill drive motor stator current signal from the calibration electrical measurement set, performs bandpass filtering on the rolling mill drive motor stator current signal to obtain a current in a frequency band of interest, applies a Hilbert transform to the current in the frequency band of interest to extract an envelope to obtain a frequency-selective current envelope sequence, calculates a spectral kurtosis value of the frequency-selective current envelope sequence, and obtains current impulse and period characteristics; The reducer state determination module integrates the amplitude and phase information of the key harmonics of the shaft power spectrum and the kurtosis value and cyclic frequency characteristics of the current impact and periodic characteristics to construct a multidimensional electrical state vector, compares each element in the multidimensional electrical state vector with a preset benchmark item by item, and outputs a rolling mill reducer operating condition determination.

2. The rolling mill reducer monitoring and diagnosis system based on multi-source information fusion according to claim 1 is characterized in that: The steps for obtaining the calibration electrical measurement set are: Through the three-phase voltage sensor and the three-phase current sensor, the continuous readings of the three-phase voltage and the three-phase current of the rolling mill drive motor are obtained in real time, and the acquisition timestamp of each reading is recorded to form the original voltage and current data; Based on the raw voltage and current data, the continuous readings of the three-phase voltage of the rolling mill drive motor and the continuous readings of the three-phase current of the rolling mill drive motor are matched one by one according to the acquisition timestamp, the time base is unified and the data are aligned to generate synchronized voltage and current data; Based on the synchronized voltage and current data, the standard reference voltage signal and the standard reference current signal are called, and the amplitude comparison and phase difference compensation are performed on the synchronized voltage and current data item by item, and the voltage amplitude and current amplitude as well as the voltage phase and current phase errors are corrected to obtain a calibrated electrical measurement set.

3. The rolling mill reducer monitoring and diagnosis system based on multi-source information fusion according to claim 1 is characterized in that: The steps for obtaining the instantaneous motor shaft power sequence are: Extracting the three-phase instantaneous voltage and current values ​​of the rolling mill drive motor based on the calibration electrical measurement set, combining the data of each phase one by one according to the phase synchronization sampling time, recording the three sets of voltage and current pairs at each sampling time in the combination, and generating a three-phase instantaneous voltage and current combination sequence; Calculating the total instantaneous input electric power at each sampling moment according to the three-phase instantaneous voltage and current combination sequence; Based on the total instantaneous input electric power at each sampling moment, the calibrated efficiency curve data of the rolling mill drive motor at different load rates is called, the corresponding sampled power value is looked up in the table to obtain the loss power value, and the loss power value is subtracted from the total instantaneous input electric power item by item to generate the instantaneous motor shaft power sequence.

4. The rolling mill reducer monitoring and diagnosis system based on multi-source information fusion according to claim 1 is characterized in that: The steps for obtaining the key harmonics of the shaft power spectrum are as follows: Based on the instantaneous motor shaft power sequence, continuous power value segments in the sequence are intercepted at fixed time intervals, and the time starting point and ending point of each segment are recorded to form a periodically segmented instantaneous shaft power segment sequence; According to the instantaneous shaft power segment sequences of the periodic segments, Fourier transform is performed on each shaft power segment sequence one by one, the shaft power value is mapped from the time domain to the frequency domain, and frequency domain power spectrum data of each shaft power segment sequence is generated; Based on the frequency domain power spectrum data, the defined target frequency points are retrieved, the amplitude value and the phase value corresponding to each target frequency point are respectively extracted from the frequency domain power spectrum data, and the key harmonics of the axis power spectrum are established.

5. The rolling mill reducer monitoring and diagnosis system based on multi-source information fusion according to claim 1 is characterized in that: The steps for obtaining the current in the frequency band of interest are: Read the amplitude of each sampling point of the stator current signal of the rolling mill drive motor from the calibration electrical measurement set, arrange them in chronological order, intercept the current amplitudes of all sampling points in the most recent 30-second period to form a sampling segment, extract the maximum amplitude, average amplitude, total number of sampling points and fixed sampling interval value in the signal segment, and generate a time-limited stator current amplitude sequence; Calculating a filter center frequency determination value based on the stator current amplitude sequence during the time limit period; Based on the filtering center frequency judgment value, the upper and lower cutoff frequencies of the bandpass filter are set with the filtering center frequency judgment value as the center, the cutoff range is set using a fixed bandwidth mode, and the filter is executed to perform bandpass processing on the frequency spectrum of the stator current amplitude sequence in the time period limit to form a frequency band current of interest.

6. The rolling mill reducer monitoring and diagnosis system based on multi-source information fusion according to claim 1 is characterized in that: The steps for obtaining the current impact and period characteristics are as follows: Extracting all sampling point data from the current in the frequency band of interest, performing Hilbert transform point by point and modulo the results to obtain the current envelope amplitude corresponding to each sampling point, forming a frequency-selective current envelope sequence, and extracting the maximum envelope amplitude, average amplitude, and original amplitude of each sampling point to form a normalized envelope amplitude structure; Calculating a spectrum kurtosis value based on the normalized envelope amplitude structure; Based on the spectrum kurtosis value, it is determined whether it is higher than the periodic impact recognition threshold. If it is higher than the periodic impact recognition threshold, it is determined that a high-amplitude impact feature exists. The periodic repeatability is then determined in combination with the uniformity of the time intervals between consecutive local peaks, and the current impact and periodic characteristics are output.

7. The rolling mill reducer monitoring and diagnosis system based on multi-source information fusion according to claim 1 is characterized in that: The steps for obtaining the multi-dimensional electrical state vector are: Calling the key harmonics of the shaft power spectrum, respectively extracting the harmonic amplitude value and the harmonic phase value corresponding to each target frequency point in the key harmonics of the shaft power spectrum, and pairing them one by one in the order of the frequency points to generate shaft power spectrum feature data including amplitude and phase; Based on the shaft power spectrum characteristic data, the current impact and period characteristics are called, the spectrum kurtosis value and the cycle frequency characteristic data in the current impact and period characteristics are extracted one by one, and time synchronization is performed with the shaft power spectrum characteristic data according to the corresponding time to generate synchronized electrical characteristic data; Based on the synchronized electrical characteristic data, the harmonic amplitude, harmonic phase, spectrum kurtosis value and cyclic frequency characteristics are arranged in order, and a characteristic parameter vector with time series correlation is constructed by combining them one by one to obtain a multidimensional electrical state vector.

8. The rolling mill reducer monitoring and diagnosis system based on multi-source information fusion according to claim 1 is characterized in that: The steps for obtaining the working condition determination of the rolling mill reducer are as follows: Extracting the value of each characteristic parameter in the multidimensional electrical state vector, respectively calling the preset reference threshold of the corresponding characteristic parameter, calculating the numerical difference between each characteristic parameter and the corresponding reference threshold one by one, and generating a difference numerical sequence between the multidimensional electrical state vector and the reference threshold; Based on the difference value sequence between the multidimensional electrical state vector and the reference threshold, determining whether the absolute value of each difference value exceeds the set fluctuation amplitude threshold one by one, recording the number and category of characteristic parameters exceeding the fluctuation amplitude threshold, and generating statistical results of the fluctuation amplitude exceeding characteristic parameters; Based on the statistical results of the fluctuation amplitude exceeding limit characteristic parameters, the number and type of the exceeding limit characteristic parameters in the statistical results are analyzed. If the number of exceeding limit characteristic parameters exceeds the preset warning threshold and the type includes periodic characteristics or harmonic characteristics, it is determined that the operating condition of the rolling mill reducer is abnormal; otherwise, the operating condition is determined to be normal, thereby forming a rolling mill reducer operating condition judgment.

Images