Centrifugal machine fault prediction system based on machine learning

Abstract

The invention relates to the technical field of fault prediction, in particular to a centrifuge fault prediction system based on machine learning, which comprises a data channel synchronization module, a multi-dimensional feature extraction module, a state evolution index construction module, a trend aggregation trajectory recognition module and a fault section recognition module. According to the method, different types of data are synchronously aligned by a multi-channel signal segmentation processing mechanism based on a periodic state, a state evolution sequence is constructed in combination with a unified sampling structure based on a time scale, a state characteristic track is established through a multi-dimensional parameter set, a trend change index is constructed by means of a difference root-mean-square between adjacent states, and the state evolution sequence is analyzed. According to the method, the aggregation section is recognized and the trajectory deviation frequency is counted by utilizing continuous trend mutation, so that dynamic migration of the trajectory boundary and intelligent recognition of the fault section are realized, the boundary failure problem caused by static preset conditions is avoided, and the continuous prediction stability of long-period equipment and the application range under a non-standard working condition are effectively enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of fault prediction, and in particular to a centrifuge fault prediction system based on machine learning. Background Art

[0002] The field of fault prediction technology mainly involves real-time monitoring and data analysis of the equipment's operating status to identify potential signs of faults and predict the time and location of their occurrence, including equipment status parameter collection, feature extraction, historical operating data analysis and life prediction modeling. It is widely used in the health management of key equipment in manufacturing, energy, transportation and other industries. Among them, the traditional centrifuge fault prediction system refers to obtaining parameters such as vibration signals, current fluctuations and speed changes of the centrifuge during operation, combining with time series-based anomaly detection methods, and comparing and analyzing the data to determine whether there are potential anomalies in the equipment. Usually, an acceleration sensor is used to collect vibration information of the equipment bearing or casing, and the original vibration signal is converted into frequency domain features through Fourier transform. The amplitude change trend within the preset frequency band is compared to identify abnormal conditions or wear conditions. At the same time, the empirical threshold judgment mechanism is combined to determine whether the equipment meets the alarm standard to indicate possible failure risks.

[0003] In the existing centrifuge fault prediction process, the frequency domain transformation of a single vibration signal is mainly relied on as the basis for fault identification. In operating scenarios with multi-source state interference or insufficient signal stability, the risk of misjudgment is prone to occur. In particular, when the centrifuge runs for a long time and produces complex conditions such as thermal deformation, current disturbance or speed fluctuation, the Fourier transform cannot accurately reveal the time evolution trend, causing the frequency band amplitude change to deviate from the actual abnormal position. At the same time, the statically set empirical threshold remains unchanged and fails to dynamically adjust with state changes. In the initial stage of parameter mutation, boundary hysteresis or premature triggering are prone to occur, making it difficult to capture trajectory changes under gradual state deviation, affecting the timeliness and continuity of equipment fault warning. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and propose a centrifuge fault prediction system based on machine learning.

[0005] In order to achieve the above objectives, the present invention adopts the following technical solutions: A centrifuge fault prediction system based on machine learning includes:

[0006] The data channel synchronization module obtains the continuous signal data of the starting point of the current spindle cycle of the centrifuge, divides the vibration analysis cycle into segments, and synchronizes the time scales of all signal segments to generate a unified segmented signal index matrix;

[0007] The multidimensional feature extraction module calculates the time difference of the peak position of the vibration signal based on the unified segmented signal index matrix, counts the proportion of the number of mutation points in the current signal, obtains the mean of the continuous fluctuation amplitude of the speed signal, calculates the temperature rise rate index of the temperature signal, and sequentially constructs a shape-forming periodic state evolution vector sequence;

[0008] The state evolution indicator construction module calculates the root mean square of the difference in each dimension as a trend change indicator based on the periodic state evolution vector sequence, records the position of the trend mutation point, and generates a trend mutation state change indicator set;

[0009] The trend aggregation trajectory identification module performs cosine similarity analysis and marks aggregation segments based on the trend mutation state change indicator set, performs trajectory deviation comparison and counts the number of boundary violations, and obtains a trajectory deviation identification sequence;

[0010] The fault section identification module calculates the difference based on the trajectory offset identification sequence, determines the boundary update amplitude and performs boundary replacement, marks the fault area, and generates a centrifuge fault prediction result.

[0011] As a further solution of the present invention, the unified segmented signal index matrix includes a periodic synchronization number, a time scale alignment identifier, a signal channel mapping relationship, a segmented time series index, and a sampling interval record; the periodic state evolution vector sequence includes a vibration position differential value, a current mutation ratio factor, a speed fluctuation mean amplitude index, a temperature rise rate change, and a time series association code; the trend mutation state change indicator set includes a trend response amplitude index, a periodic fluctuation mean square error value, an abnormal evolution position index, and an inter-vector difference expression degree; the trajectory offset identifier sequence includes a similarity aggregation label, a direction consistency count, a trajectory change interval envelope value, and an out-of-bounds period position record; the centrifuge fault prediction result includes a predicted state segment number, a fault evolution path label, a trajectory center difference metric, and a fault level discrimination result.

[0012] As a further solution of the present invention, the data channel synchronization module includes:

[0013] The signal acquisition submodule acquires continuous signal data from the vibration acceleration sensor, axial speed recorder, current monitoring channel, and temperature sensing device. Based on the current spindle cycle starting point as a reference, the acquisition reference points are used to divide the vibration analysis cycle, and all signal data segments are collected to generate a multi-source synchronous signal segment set.

[0014] The time synchronization submodule reads the timestamp data inside each signal segment based on the multi-source synchronization signal segment set, calculates the time deviation between each signal segment and the starting point of the spindle cycle, adjusts the time axis data of each signal segment based on the time reference of the starting point of the spindle cycle, compares whether the deviation exceeds the time synchronization threshold, and if so, adds or subtracts the signal time axis until the time axis of each signal segment meets the synchronization standard, thereby obtaining a unified time-aligned signal segment set;

[0015] The signal index generation submodule records the start and end time points, signal length, sampling rate and channel sequence number of each signal segment according to the unified time-aligned signal segment set, combines the time points and channel numbers of each signal segment with the spindle cycle starting point reference benchmark, establishes a mapping index between the signal segment and the spindle cycle reference point, and generates a unified segmented signal index matrix.

[0016] As a further solution of the present invention, the multidimensional feature extraction module includes:

[0017] The vibration feature calculation submodule obtains the vibration acceleration signal data segments based on the unified segmented signal index matrix, selects all vibration data points in each signal segment, detects the vibration amplitude point by point, records the time positions of all peak points, calculates the time difference between each pair of adjacent peak points, and summarizes the vibration signal peak time difference;

[0018] The current and speed feature extraction submodule reads the current signal data segment based on the time difference of the vibration signal peak value, performs point-by-point amplitude difference calculation on all sampling points, determines whether the differential data exceeds the set current mutation threshold, counts the number of mutation points, and simultaneously calculates the proportion of the number of current signal mutation points. It also performs absolute value calculation on the speed data of consecutive sampling points, calculates the mean of the fluctuation amplitude, and obtains the current mutation proportion and the mean of the speed fluctuation.

[0019] The temperature feature analysis submodule reads the temperature signal data segment based on the current mutation ratio and the speed fluctuation mean, performs point-by-point differentiation on the temperature values ​​of all sampling points, calculates the temperature rise rate per unit time based on the time interval, and obtains the average temperature rise rate index through averaging. Combined with the vibration signal peak time difference, the current mutation ratio and the speed fluctuation mean, the feature results are sorted in chronological order, and combined according to the time series to establish a periodic state evolution vector sequence.

[0020] As a further solution of the present invention, the state evolution indicator construction module includes:

[0021] The difference extraction submodule sequentially obtains three adjacent vector groups based on the periodic state evolution vector sequence, extracts the values ​​of each group of three vectors in the same dimension, performs a subtraction operation on the values ​​between two adjacent periods in the same dimension, repeats the operation for all dimensions, and obtains a periodic state difference sequence table;

[0022] The trend calculation submodule performs a square operation and an average root operation on the difference items of each dimension according to the period state difference sequence table to obtain the root mean square trend change value under the corresponding dimension. Combined with the previous and next data of the current period, the trend change intensity is calculated to obtain the trend change intensity sequence;

[0023] The mutation judgment submodule sets a fault diagnosis threshold based on the trend change intensity sequence, performs a threshold comparison operation on the trend intensity values ​​of three consecutive cycles, and judges one by one whether the three-cycle sequences are all greater than the fault diagnosis threshold. If the conditions are met, the current cycle number is recorded as the position of the trend mutation point, and the state evolution data corresponding to all mutation points are extracted and sorted to establish a trend mutation state change indicator set.

[0024] As a further solution of the present invention, the trend aggregation trajectory identification module includes:

[0025] The directional consistency identification submodule obtains the state vectors of the two cycles before and after each mutation point based on the trend mutation state change indicator set, extracts the vector data of each dimension of the cycle, calculates the cosine value of the angle, and determines whether the cosine value is greater than or equal to the directional consistency benchmark threshold. If the condition is met, it is marked as a directional consistent dimension, otherwise it is marked as a directional inconsistent dimension, and a sequence of directional consistent dimension values ​​is obtained;

[0026] The aggregation segment determination submodule obtains the direction-consistent dimension value of each cycle based on the direction-consistent dimension number value sequence, determines whether the direction-consistent dimension number of each cycle is greater than or equal to the principal component contribution rate threshold, and if the condition is met, marks the corresponding cycle location as an aggregation cycle, identifies the continuous cycle number range and sets it as an aggregation segment, compares the minimum and maximum values ​​of each dimension state value, obtains the upper and lower boundary values ​​of the state within the segment, and obtains the state envelope trajectory interval;

[0027] The trajectory deviation statistics submodule judges the state vector of each subsequent cycle in turn according to the state envelope trajectory interval, compares it with the maximum and minimum values ​​of the envelope interval of the corresponding dimension, and determines whether it exceeds the interval boundary range. If the dimension state value is greater than the maximum value or less than the minimum value, the corresponding dimension is marked as an out-of-bounds dimension. The number of out-of-bounds dimensions in each cycle is counted and the out-of-bounds ratio is calculated. If it is greater than the trajectory deviation judgment reference value, the cycle state is marked as a trajectory deviation, and a trajectory deviation identification sequence is generated.

[0028] As a further solution of the present invention, the fault section identification module includes:

[0029] The cross-border accumulation judgment submodule traverses and judges all cycles based on the trajectory offset identification sequence, reads the value of each cycle in the offset identification value sequence, and records the last cycle of the segment as the cross-border judgment trigger point if there are more than three consecutive cycles. The numbers of all cycles that meet the cross-border judgment number of more than three consecutive times are counted to obtain the trigger cycle sequence;

[0030] The boundary update decision submodule reads the state vectors of the five cycles before each trigger cycle according to the trigger cycle sequence, calculates the five-cycle mean for each dimension, obtains the upper and lower boundaries of the aggregation segment state envelope trajectory interval, calculates the dimension difference strength, records the number of update marks, and uses the current mean interval as the update boundary after the cumulative number reaches three times to form a corrected state boundary interval;

[0031] The fault section marking submodule performs tracking judgment on the period in which the offset continues to occur after the update based on the correction state boundary interval and the trajectory offset identification sequence. It detects whether the continuous offset is still in the out-of-bounds state after completing three boundary updates. If it meets the requirements, the corresponding period is marked as a fault section period, the fault period number is output, and the centrifuge fault prediction result is obtained.

[0032] Compared with the prior art, the advantages and positive effects of the present invention are:

[0033] In the present invention, a multi-channel signal segmentation processing mechanism based on periodic states synchronously aligns different types of data, and combines a unified sampling structure based on time scales to construct a state evolution sequence. The state characteristic trajectory is established by extracting a parameter set including multiple dimensions such as time displacement difference, current mutation intensity, fluctuation amplitude and temperature rise change. The trend change index is constructed based on the root mean square difference between adjacent states. The continuous trend mutation is used to identify the aggregation segment and the trajectory offset frequency is counted to realize the dynamic migration of the trajectory boundary and the intelligent identification of the fault segment, avoiding the boundary failure problem caused by static preset conditions, constructing real-time judgment logic for state evolution, solving the problem of insufficient model adaptability under operating cycle changes, and effectively enhancing the continuous prediction stability of long-cycle equipment and the adaptability range under non-standard working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is a system flow chart of the present invention;

[0035] Figure 2 This is a flow chart of the data channel synchronization module of the present invention;

[0036] Figure 3 This is a flow chart of the multidimensional feature extraction module of the present invention;

[0037] Figure 4 A flow chart of the module for constructing the state evolution indicator of the present invention;

[0038] Figure 5 This is a flow chart of the trend aggregation trajectory identification module of the present invention;

[0039] Figure 6 This is a flow chart of the fault section identification module of the present invention. DETAILED DESCRIPTION

[0040] 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.

[0041] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0042] See also Figure 1 , a centrifuge fault prediction system based on machine learning includes:

[0043] The data channel synchronization module acquires continuous signal data from the vibration acceleration sensor, axial speed recorder, current monitoring channel, and temperature sensing device deployed in the centrifuge equipment after the start of the current spindle cycle. It divides the vibration analysis cycle (30 seconds) into segments based on the cycle time starting point as a reference, and synchronizes the time scale of all signal segments (implementing the precise time protocol) to generate a unified segmented signal index matrix.

[0044] The multidimensional feature extraction module uses a unified segmented signal index matrix to calculate the time difference between the peak positions of the vibration signal, count the percentage of mutation points in the current signal, obtain the mean of the continuous fluctuation amplitude of the speed signal, calculate the temperature rise rate index of the temperature signal (in compliance with the temperature rise test standard for high-voltage switchgear), and sequentially construct the state time series to form a periodic state evolution vector sequence.

[0045] The state evolution indicator construction module is based on the periodic state evolution vector sequence. It sequentially obtains the difference between the values ​​of each dimension between three sets of adjacent vectors and calculates the root mean square of the difference in each dimension as a trend change indicator. When three consecutive cycles exceed the fault diagnosis threshold, it is recorded as the trend mutation point position, and a trend mutation state change indicator set is generated.

[0046] The trend aggregation trajectory identification module uses a set of trend mutation state change indicators to perform cosine similarity analysis (threshold set to ≥0.85) on the state vectors of the two cycles before and after each mutation point to determine whether the change directions of each dimension are consistent. If the number of dimensions with consistent directions exceeds the principal component contribution rate threshold (set to 70% of the total dimensions), it is marked as an aggregation segment, and the corresponding change amplitude interval is calculated to form a state envelope trajectory. The trajectory offset comparison is performed on the state vector of each subsequent cycle, and the number of cross-borders is counted to obtain a trajectory offset identification sequence.

[0047] The fault section identification module is based on the trajectory offset identification sequence. If the number of consecutive boundary violations exceeds three, the difference between the mean of the state vector of the five cycles before each boundary violation point and the center point of the aggregated trajectory is calculated, the boundary update amplitude is determined, and boundary replacement is performed. If the update accumulates to three times and the boundary is still in the out-of-bounds state, it is marked as a fault section and the centrifuge fault prediction result is generated.

[0048] The unified segmented signal index matrix includes periodic synchronization number, time scale alignment identifier, signal channel mapping relationship, segmented time series index, and sampling interval record; the periodic state evolution vector sequence includes vibration position differential value, current mutation ratio factor, speed fluctuation mean amplitude index, temperature rise rate change, and time series association code; the trend mutation state change indicator set includes trend response amplitude index, periodic fluctuation mean square error value, abnormal evolution position index, and inter-vector difference expression degree; the trajectory offset identifier sequence includes similarity aggregation label, direction consistency count, trajectory change interval envelope value, and out-of-bounds period position record; the centrifuge fault prediction result includes predicted state segment number, fault evolution path label, trajectory center difference metric, and fault level judgment result.

[0049] See also Figure 2 , the data channel synchronization module includes:

[0050] The signal acquisition submodule acquires continuous signal data from the vibration acceleration sensor, axial speed recorder, current monitoring channel, and temperature sensing device. Based on the current spindle cycle starting point as a reference, the acquisition reference points are used to divide the vibration analysis cycle, and all signal data segments are collected to generate a multi-source synchronous signal segment set.

[0051] Obtain continuous signal data from the vibration acceleration sensor, axial speed recorder, current monitoring channel, and temperature sensing device. First, test the vibration acceleration channel and select the CA-YD102 triaxial vibration acceleration sensor. Set the sampling frequency to 1024Hz and acquire real-time data through the PLC acquisition module. Collect 1024 data points per second and set the sampling time to 30 seconds. The total number of acquisition points is 1024×30=30720 points. After being cached in the local buffer, it is written into the industrial database. Then monitor the axial speed recorder and select the ZX-RPM300 series. The recorder has a sampling frequency of 256Hz, and the total number of sampling points in 30 seconds is 256×30=7680 points. The data is saved in CSV format. Then, the signal of the current monitoring channel is acquired. A current transmitter with a rated accuracy of ±0.5%FS is selected. The sampling frequency is set to 100Hz. The number of sampling points in 30 seconds is 100×30=3000 points. The CAN bus protocol is used for data upload. Finally, the temperature sensing device is collected. A Pt100 thermal resistor temperature sensor is used. The sampling frequency is 10Hz. The number of sampling points in 30 seconds is 10×30=300 points. Modbus is used. The RTU communication protocol is uploaded to the control system server in real time. After completing the acquisition of the above four channel signal data, the software module of the data receiving end is used to distinguish and store all signal data according to channel type, and each signal segment is marked with the start time, sampling frequency, and channel number. The data is integrated into a unified data format to form a multi-channel and multi-time series structure. Combined with the time mark of the current spindle cycle starting point, all signal data segments are numbered and grouped according to the start and end time periods to obtain a multi-source synchronous signal segment set. The specific sampling parameters are shown in Table 1.

[0052] Table 1 Signal acquisition parameters:

[0053] As shown in Table 1, the specific sampling parameters of each signal channel are listed.

[0054] The time synchronization submodule reads the timestamp data within each signal segment based on the multi-source synchronization signal segment set, calculates the time deviation between each signal segment and the start point of the spindle cycle, and adjusts the time axis data of each signal segment based on the time reference of the spindle cycle start point. It compares whether the deviation exceeds the time synchronization threshold. If so, it adds or subtracts the signal time axis until the time axis of each signal segment meets the synchronization standard, thus obtaining a unified time-aligned signal segment set.

[0055] According to the multi-source synchronous signal segment set, the timestamp data recorded in each signal segment is called. First, the starting timestamp data of the vibration acceleration channel is extracted, and the reading starting time is 0.0025 seconds. Compared with the starting time of the spindle cycle 0 seconds, the time deviation is 2.5 milliseconds. Then, the time deviation of the axial speed signal segment is detected. The starting time is -0.003 seconds and the deviation is -3.0 milliseconds. Next, the time deviation of the current monitoring channel is measured. The starting time is 0.0018 seconds and the deviation is 1.8 milliseconds. Finally, the temperature sensing signal segment is detected. The starting time is -0.0022 seconds and the deviation is -2.2 milliseconds. The synchronization threshold is set to 5 milliseconds. It is judged whether the time deviation of each signal segment is greater than or less than the threshold. According to the deviation The absolute value is compared with the 5-ms threshold. The deviations of all signal segments meet the threshold condition. The time axis is shifted left for the signal segments with deviations greater than zero. The shift time is the corresponding deviation. For example, the start time of the vibration acceleration signal segment is adjusted from 0.0025 seconds to 0 seconds. The time axis is shifted right for the signal segments with deviations less than zero. The start time of the axial speed signal segment is adjusted from -0.003 seconds to 0 seconds, the current monitoring signal segment is adjusted from 0.0018 seconds to 0 seconds, and the temperature sensing signal segment is adjusted from -0.0022 seconds to 0 seconds. After the adjustment, the start time of the four signal segments is re-checked to confirm that the start time of all signal segments is aligned with the spindle cycle starting point of 0 seconds, and a unified time-aligned signal segment set is obtained. The relevant deviation adjustment data is shown in Table 2.

[0056] Table 2 Time synchronization adjustment table:

[0057] Table 2 shows the time deviation detection and adjustment records for each signal segment.

[0058] The signal index generation submodule aligns the signal segments according to the unified time, records the start and end time points, signal length, sampling rate and channel sequence number of each signal segment, combines the time points and channel numbers of each signal segment with the reference benchmark of the spindle cycle starting point, establishes the mapping index between the signal segment and the spindle cycle reference point, and generates a unified segmented signal index matrix;

[0059] According to the unified time alignment signal segment set, the start and end time points, data length, sampling rate and channel serial number of the four signal segments of vibration acceleration, axial speed, current monitoring and temperature sensing are extracted. First, the vibration acceleration signal segment is numbered, and the channel number is set to 1, the corresponding time period is 0 milliseconds to 30000 milliseconds, the data length is 30720 points, and the index serial number is set to 101. Then the axial speed signal segment is numbered 2, the corresponding time period is also 0 milliseconds to 30000 milliseconds, the data length is 7680 points, the index serial number is 102, and the current monitoring signal segment is numbered 3. The time segment is from 0 milliseconds to 30000 milliseconds, the data length is 3000 points, and the index number is 103. The temperature sensing signal segment number is 4, the time segment is from 0 milliseconds to 30000 milliseconds, the data length is 300 points, and the index number is 104. By reading the channel number and time period information of each signal segment, the signal segments are sorted according to the ascending rule of the channel number, and a mapping relationship is established according to the order of the signal segment numbers. The signal segment number, corresponding time period, channel number and index number are integrated to form a signal segment index data set with four rows and one column, and finally a unified segmented signal index matrix is ​​generated, as shown in Table 3.

[0060] Table 3 Signal index mapping table:

[0061] Table 3 shows the detailed mapping data between signal segments and corresponding index numbers.

[0062] See also Figure 3 , the multi-dimensional feature extraction module includes:

[0063] The vibration feature calculation submodule obtains the vibration acceleration signal data segments based on the unified segmented signal index matrix, selects all vibration data points in each signal segment, detects the vibration amplitude point by point, records the time positions of all peak points, calculates the time difference between each pair of adjacent peak points, and summarizes the vibration signal peak time difference;

[0064] Obtain a unified segmented signal index matrix, extract all data segments of the vibration acceleration signal channel in turn, and perform point-by-point vibration amplitude detection on each data segment. First, read the vibration signal data with a sampling frequency of 1024Hz, and assign 1024 data points collected per second to the time axis. Use a threshold of 0.8g for peak detection to determine whether each sampling point meets the peak condition. When the vibration amplitude exceeds 0.8g, it is identified as a peak point. For example, the vibration amplitude of the 125th point is 1.2g, the vibration amplitude of the 280th point is 1.1g, and the vibration amplitude of the 430th point is 1.3g. Extract the timestamps of these peak points Data, the time difference of all adjacent peak pairs is calculated. For example, the corresponding time of point 125 is 0.122 seconds, and that of point 280 is 0.273 seconds, with a time difference of 0.151 seconds between the two points. The time differences of all subsequent peak point pairs are solved in sequence, and the number of peak pairs is counted. Assuming that a total of 150 pairs of peak points are detected in a 30-second data segment, the total accumulated time difference is 1.86 seconds. The arithmetic average method is used to divide the total time difference by the number of peak pairs, that is, 1.86 / 150, to obtain an average peak time difference of 0.0124 seconds, which is converted to 12.4 milliseconds. Finally, the peak time difference of the vibration signal is generated, as shown in Table 4.

[0065] Table 4 Vibration peak time difference calculation table:

[0066] Table 4 shows the calculation process data of the time difference of the vibration signal peak pairs in the embodiment.

[0067] The current and speed feature extraction submodule reads the current signal data segment based on the time difference of the vibration signal peak, performs point-by-point amplitude difference calculation on all sampling points, determines whether the differential data exceeds the set current mutation threshold, counts the number of mutation points, and simultaneously calculates the proportion of the number of current signal mutation points. It also performs absolute value calculation on the speed data of consecutive sampling points, calculates the mean fluctuation amplitude, and obtains the current mutation proportion and the speed fluctuation mean.

[0068] According to the time difference of the vibration signal peak, the current signal data segment is extracted, and the current differential operation is performed point by point on all adjacent sampling points. The current sampling frequency is set to 100Hz and the sampling time interval is 0.01 seconds. For each pair of adjacent sampling points, the current change is calculated, and the threshold ±2A is used to determine the mutation point. For example, if the current at the 100th point is 8.0A and the current at the 101st point is 10.8A, the difference value is 2.8A, which is determined to be a mutation point. The number of mutation points is accumulated. If 150 mutation points are detected in 3000 sampling points, the number of mutation points accounts for 150 / 3000=0.05, that is, 5%. , then call the speed signal data segment, read the speed signal with a sampling frequency of 256Hz, and perform absolute value processing on the speed change between adjacent sampling points. For example, the speed at the 500th point is 1490rpm, and the speed at the 501st point is 1495rpm. The single-point change is 5rpm. In this way, the 7680-point data is calculated pair by pair. After summing up all the changes, the total fluctuation change is 24576rpm. Then, divided by the number of sampling points (7680), the average fluctuation amplitude is calculated as 24576 / 7680=3.2rpm. Finally, the current mutation ratio and the speed fluctuation average are obtained, as shown in Table 5.

[0069] Table 5 Current and speed characteristic calculation table:

[0070] As shown in Table 5, the calculation process data of the current mutation ratio and the speed fluctuation mean are shown.

[0071] The temperature feature analysis submodule reads the temperature signal data segment based on the current mutation ratio and the mean speed fluctuation, performs point-by-point differentiation on the temperature values ​​of all sampling points, calculates the temperature rise rate per unit time based on the time interval, and averages the results to obtain the average temperature rise rate indicator. The results are then sorted chronologically based on the vibration signal peak time difference, the current mutation ratio, and the mean speed fluctuation, and combined in time series to establish a periodic state evolution vector sequence.

[0072] Based on the current mutation ratio and the speed fluctuation mean, the temperature signal data segment was extracted. The temperature sampling data with a sampling frequency of 10 Hz and a sampling interval of 0.1 second were read. The temperature difference calculation was performed for each pair of adjacent sampling points. For example, the temperature at point 50 was 25.0°C and the temperature at point 51 was 25.3°C, with a temperature difference of 0.3°C and a temperature rise rate of 0.3 / 0.1=3.0°C / s. The temperature rise rate was calculated for all 299 pairs of temperature points in the segment. The total temperature rise rate was calculated to be 450°C / s. Dividing it by 299 pairs of points, the average temperature rise rate was 450 / 299=1.51°C / s. Then, combined with the vibration signal peak time difference of 12.4 ms, the current mutation ratio and the speed fluctuation mean of 0.05 and 3.2 rpm, and the temperature rise rate of 1.51°C / s, the data were arranged in chronological order. The characteristic values ​​were organized into a four-dimensional vector sequence to represent the state evolution process under the corresponding period. Finally, the periodic state evolution vector sequence was established, as shown in Table 6.

[0073] Table 6 Cycle state evolution vector table:

[0074] As shown in Table 6, the specific characteristic quantities of the periodic state evolution vector sequence are summarized.

[0075] See also Figure 4 , the state evolution indicator building blocks include:

[0076] The difference extraction submodule obtains three adjacent vector groups in sequence based on the periodic state evolution vector sequence, extracts the values ​​of each group of three vectors in the same dimension, performs a subtraction operation on the values ​​between two adjacent periods in the same dimension, repeats the operation for all dimensions, and obtains a periodic state difference sequence table;

[0077] Based on the periodic state evolution vector sequence, the state vector samples of three adjacent periods are extracted in turn, and the values ​​of the four dimensions (vibration signal, current signal, speed signal, temperature signal) in each group of vectors are extracted dimension by dimension, and the difference operation is performed on the same dimension values ​​between adjacent periods. For example, between the first and second periods, the vibration dimension values ​​are 12.4ms and 11.7ms, with a difference of 0.7ms, the current mutation proportions are 0.08 and 0.05, respectively, with a difference of 0.03, the speed fluctuation mean is 3.2rpm and 7.2rpm, with a difference of 4rpm, the temperature rise rate is 1.5℃ / s and 1.2℃ / s, with a difference of 0.3℃ / s, and all dimensions are calculated. The same operation is used to record the obtained difference data. At the same time, the cycle number is traversed in a traversal manner to ensure that each comparison uses adjacent cycle data for difference. For example, after repeating the above operation for the second and third cycles, a vibration difference of 1.1ms, a current difference of 0.02, a speed difference of 5rpm, and a temperature rise difference of 0.4℃ / s are obtained. After processing the third and fourth cycles, a vibration difference of 0.5ms, a current difference of 0.06, a speed difference of 3rpm, and a temperature rise difference of 0.5℃ / s are obtained. On this basis, a multidimensional difference data set is formed, and the difference values ​​of each cycle are numbered and established into a structured table. The results are shown in Table 7, and a periodic state difference sequence table is obtained.

[0078] Table 7 Example of periodic state difference:

[0079] As shown in Table 7, the four-dimensional characteristic differences between adjacent weeks are recorded for subsequent trend strength indicator calculation.

[0080] The trend calculation submodule performs square operations and averages the difference items of each dimension according to the period state difference sequence table to obtain the root mean square trend change value under the corresponding dimension, combines the data before and after the current period, and uses the formula:

[0081] ;

[0082] The trend change intensity is obtained by operation, and the trend change intensity sequence is obtained, where: Indicates the strength of trend change, the unit is the difference The same unit of change, Indicates the The difference between the state vector of the current cycle and the state vector of the previous cycle in dimensions, Represents the average value of all dimension differences, Indicates the The trend change gradient of the dimension, Indicates the The value of the change amplitude of the dimension state, the unit is same, Indicates the deviation value of the cycle state change trend, the unit is consistent with the denominator;

[0083] According to the periodic state difference sequence table, first count the difference sets in each dimension, such as the vibration difference is [0.7, 1.1, 0.5], the current difference is [0.03, 0.02, 0.06], the speed difference is [4, 5, 3], and the temperature rise difference is [0.3, 0.4, 0.5]. Calculate the average value of the difference sequence of each dimension respectively, and the obtained vibration average difference is ms, the average current difference is 0.037, the current change is small and does not require normalization, the average speed is 4.0rpm, and the average temperature rise is 0.4℃ / s. Then, taking the vibration dimension as an example, a trend gradient indicator sequence is constructed, which is defined as the rate of change of continuous period difference, which can be expressed as , where the periodic sampling interval is 1 unit period, then the G sequence is [0.4, -0.6], etc., and then the trend strength value is calculated according to the formula:

[0084] Bring in a numerical example, assuming:

[0085] The vibration dimension difference sequence is ;

[0086] mean ;

[0087] Gradient value sequence ;

[0088] State change amplitude sequence ;

[0089] Trend Deviation ;

[0090] Calculate in sequence:

[0091] The numerator is:

[0092] ;

[0093] The denominator is:

[0094] ;

[0095] The trend change strength values ​​are:

[0096] ;

[0097] The results show that the trend change intensity is -0.0346, and finally the trend change intensity sequence is obtained.

[0098] Trend change intensity is a comprehensive measurement indicator used to measure the changing trend of the periodic state vector in a continuous time series. It reflects the degree of coupling between the fluctuation amplitude and the change rate of the system relative to the average state in multiple characteristic dimensions. This indicator not only takes into account the difference amplitude of each dimension between the current cycle and the adjacent cycle, but also introduces the dynamic parameter of trend gradient, thereby realizing the quantitative assessment of "whether the change continues to intensify". At the same time, by integrating the multi-dimensional state amplitude fluctuation and the degree of periodic offset, a normalized benchmark is constructed, so that it can maintain dimensional consistency and comparability under different scales and units. Therefore, the trend change intensity not only reflects the severity of the state change, but also reflects the coordination and mutation between the changes in each dimension. It is the key basis for subsequent judgment of the location of the trend mutation point.

[0099] The numerator of the formula is calculated by taking the current period difference of each dimension The average difference from all dimensions The absolute value of the difference is multiplied by the trend change gradient of the corresponding dimension , achieving dual weighting of the degree of difference and the rate of change, where the absolute value operation is used to remove the directional influence, so that the formula focuses on the size of the change rather than the positive and negative fluctuations, the multiplication operation reflects the dynamic amplification or attenuation of the difference value under the trend change, and the summation operation uniformly accumulates the trend difference effects of all dimensions to form the global trend driving total; the denominator is composed of the state change amplitude of all dimensions The square sum of the energy index is used to restore the original dimension of the change and the deviation degree of the periodic trend. The absolute value of is added to construct the total resistance term of the trend disturbance, where the sum of squares reflects the current scale of the overall change of the system. The square root operation makes it comparable with the numerator term in the same dimension. The absolute value term It is the disturbance factor introduced by irregular fluctuations across cycles, and its introduction constitutes a dynamic adjustment factor of the denominator to the trend strength. The overall formula thus structurally forms the ratio between the "trend incentive amount (numerator)" and the "system blockage amount (denominator)", and has a multi-dimensional trend strength determination function that is direction-independent, amplitude-weighted, and has controllable disturbances.

[0100] The mutation judgment submodule sets a fault diagnosis threshold based on the trend change strength sequence, performs a threshold comparison operation on the trend strength values ​​of three consecutive cycles, and judges whether the three-cycle sequence is greater than the fault diagnosis threshold one by one. If the condition is met, the current cycle number is recorded as the position of the trend mutation point, and the state evolution data corresponding to all mutation points are extracted and sorted to establish a trend mutation state change indicator set;

[0101] According to the trend change intensity sequence, the judgment operation is performed on the continuous period T value sequence, and the trend mutation threshold is defined as 3.5. The threshold setting is based on the 95% confidence upper limit of the state change intensity during system operation. In large-scale operation condition monitoring, after counting 1000 groups of different period state data, it is observed that the average amplitude of the difference in all dimensions does not exceed 0.9, the gradient value does not exceed ±0.6, and the period trend deviation degree is less than 0.9. Maintained between 0.3 and 0.6, under this combination of conditions, the T value is concentrated in the range of 0.5 to 3.2. Therefore, the upper bound interval is expanded with a margin of 0.1 as the critical point for mutation identification and is set to 3.5. This value is most sensitive to the fluctuation intensity, especially with the increase of vibration difference and temperature rise fluctuation, it shows a significant upward trend and has a quantifiable judgment effect. The judgment condition is that the trend intensity values ​​of three consecutive cycles exceed the threshold. For example, a sequence T=[3.2, 3.7, 4.0, 3.6, 3.3, 3.8] is used to detect whether the sequence segments meet the conditions in a sliding window manner. The T value in the 2nd to 4th cycle interval is The values ​​of [3.7, 4.0, 3.6] are [3.7, 4.0, 3.6], all exceeding the threshold of 3.5 and satisfying the mutation condition. Therefore, cycle 4 is recorded as the trend mutation point position. All cycle numbers that meet the conditions are recorded, the number of cycle mutation points is counted, and the corresponding original state feature vector data is extracted. That is, the original state sequence of cycle 4 is vibration time difference 12.8ms, current mutation ratio 0.085, speed fluctuation 3.9rpm, and temperature rise rate 1.65℃ / s. The multidimensional state vector data corresponding to this cycle is used as the mutation state mark sample. All mutation point samples are repeatedly recorded to finally establish a trend mutation state change indicator set.

[0102] See also Figure 5 ,The trend aggregation trajectory identification module includes:

[0103] The directional consistency identification submodule obtains the state vectors of the two cycles before and after each mutation point based on the trend mutation state change indicator set, extracts the vector data of each dimension of the cycle, calculates the cosine value of the angle, and determines whether the cosine value is greater than or equal to the directional consistency benchmark threshold. If the condition is met, it is marked as a directional consistent dimension, otherwise it is marked as a directional inconsistent dimension, and the number value sequence of the directional consistent dimension is obtained;

[0104] Based on the trend mutation state change indicator set, extract all mutation point cycle numbers and obtain the state vectors of the two cycles before and after each mutation cycle. Let the mutation point cycle be , extraction cycle and The four dimensional state values ​​are: vibration dimension: , current dimension: , speed dimension: , temperature rise dimension: , the state value pairs of corresponding dimensions are respectively formed into vectors and , the cosine similarity calculation method is used to judge the direction, and the dot product calculation is to multiply the components of each dimension and then sum them. For example, the dot product of the current dimension is , the module lengths are , the module length product is , the cosine value is , the direction consistency benchmark threshold is set to This value is obtained by collecting 256 mutation cycle points and their corresponding state vectors before and after the cycle in a single day of full-day operation, and calculating the cosine value distribution of each dimension. The results show that the cosine values ​​in more than 99% of the normal operating cycle segments are concentrated in the range of 0.88 to 0.99, while the cosine values ​​of most disturbance dimensions in all state mutation cycles are lower than 0.80. Therefore, 0.85 is selected as the dividing point to exclude normal weak fluctuations and cover the direction reversal of common mutation dimensions. At the same time, the threshold will decrease slightly as the range of violent fluctuations in the vibration dimension and the speed dimension increases. The empirical relationship, Represents the average value of the multi-dimensional state variance. Determine whether the cosine value is greater than the threshold. If the condition is met, the direction of the dimension is consistent. Repeat this process in each dimension. The direction consistency of each dimension in cycle 3 is as follows:

[0105] Table 8 Periodic state vector direction consistency determination table:

[0106] Repeat the above process for each period, count the number of dimensions with consistent directions, and get the number of dimensions with consistent directions in period 3 as , and then obtain a sequence of dimension values ​​with consistent direction.

[0107] The aggregation segment determination submodule obtains the direction-consistent dimension value of each cycle based on the direction-consistent dimension number value sequence, and determines whether the direction-consistent dimension number of each cycle is greater than or equal to the principal component contribution rate threshold. If the condition is met, the corresponding cycle position is marked as an aggregation cycle, and the continuous cycle number range is identified and set as an aggregation segment. The minimum and maximum values ​​of each dimension state value are compared to obtain the upper and lower boundary values ​​of the state within the segment, and the state envelope trajectory interval is obtained;

[0108] According to the direction of the number of dimensions , get the number of dimensions with consistent direction in each cycle, and the system sets the total number of dimensions , the principal component contribution rate threshold ratio is set to , calculate the aggregation judgment threshold as , round up to get ,like Then the cycle Marked as aggregation period, for example, period 4 has a consistent dimension of 6, and period 5 has a consistent dimension of 7, forming a continuous segment , extract all periodic state vectors of this segment, perform upper and lower boundary extraction operations on each dimension, and the vibration dimension values ​​are , the current dimension is , the speed dimension is , the temperature rise dimension is , construct the envelope interval as follows:

[0109] Table 9 Aggregation segment state envelope trajectory interval table:

[0110] Finally obtain the state envelope trajectory interval ,in is the dimension number, is the lower boundary, is the upper boundary.

[0111] The trajectory deviation statistics submodule judges the state vector of each subsequent cycle in turn according to the state envelope trajectory interval, compares it with the maximum and minimum values ​​of the envelope interval of the corresponding dimension, and determines whether it exceeds the interval boundary range. If the dimension state value is greater than the maximum value or less than the minimum value, the corresponding dimension is marked as an out-of-bounds dimension. The number of out-of-bounds dimensions in each cycle is counted and the out-of-bounds ratio is calculated. If it is greater than the trajectory deviation judgment reference value, the cycle state is marked as a trajectory deviation, and a trajectory deviation identification sequence is generated.

[0112] According to the state envelope trajectory interval , for the cycle The state vector Execute the judgment, if the state value of a dimension It is considered out of bounds, and the state vector in period 6 is: vibration , current , speed , temperature rise , then the out-of-bounds dimensions are vibration and current; the number of out-of-bounds dimensions is 2, accounting for , does not trigger trajectory deviation; the state in cycle 7 is: vibration , current , speed , temperature rise , all 4 dimensions are out of bounds, the number of out-of-bounds , accounting for , meet the trajectory deviation judgment threshold, record cycle 7 as the trajectory deviation cycle, judge cycle 8 in turn and summarize the cycle deviation situation to generate the trajectory deviation identification sequence ,in Represents a period offset, Indicates no offset.

[0113] See also Figure 6 , the fault section identification module includes:

[0114] The out-of-bounds cumulative judgment submodule traverses and judges all cycles based on the trajectory offset identification sequence, reads the value of each cycle in the offset identification value sequence, and records the last cycle of the segment as the out-of-bounds judgment trigger point if there are more than three consecutive cycles. The numbers of all cycles that meet the requirement of more than three consecutive out-of-bounds are counted to obtain the trigger cycle sequence;

[0115] Based on the offset status identification value in the trajectory offset identification sequence, read the offset flag corresponding to each cycle in the sequence If the cycle satisfies the condition that four consecutive cycles are all 1, the last cycle is recorded as the trigger point for out-of-bounds judgment. First, obtain the offset flag value of the entire sequence. In the sample, the offset flag values ​​of cycles 2 to 5 are all 1, which meets the condition that four consecutive cycles are out of bounds. Therefore, cycle 5 is marked as the trigger cycle. The sample data is continued to judge the values ​​of cycles 6 to 9. Among them, cycles 7 to 9 have three consecutive cycles of offset, which does not meet the recording condition, and the judgment is terminated. In actual operation, it is necessary to perform a sequence of length The offset identification array performs a sliding window operation to determine whether each sub-array of length 4 is all 1, and extracts the window segment through a loop And perform logical judgment operations , each time it is satisfied, Adding a trigger period set, Table 10 provides a set of sample data:

[0116] Table 10: Example of trajectory offset identification sequence:

[0117] As shown in Table 10, the system identifies a single window that meets the conditions, corresponding to the trigger cycle number "Cycle 5." In a real-world scenario, this operation can process a state sequence with a sampling period of 0.5 seconds, indicating that the system is always in an abnormal offset state for 2 consecutive seconds, ultimately resulting in a trigger cycle sequence.

[0118] The boundary update decision submodule reads the state vectors of the five cycles before each trigger cycle according to the trigger cycle sequence, calculates the five-cycle mean for each dimension, and obtains the upper and lower boundaries of the state envelope trajectory interval of the aggregation segment using the formula:

[0119] ;

[0120] The operation obtains the dimension difference strength, records the number of update marks, and when the cumulative number reaches three, uses the current mean interval as the update boundary to form the correction state boundary interval, where, Represents a period In dimension The normalized difference intensity on Represents a period First 5 cycles in dimension The state mean on , For the Cycle in dimension The status value on Aggregate segments in dimensions The lower and upper boundaries of the state envelope, Indicates the fluctuation span of the state trajectory of this dimension, which is used for normalization processing;

[0121] According to the trigger cycle sequence obtained above, for each trigger cycle Perform status traceback operations to obtain cycles arrive The state vector of , extracts five state values ​​for each dimension and calculates their mean , then call the preset state envelope trajectory boundary , find the center point of the boundary and interval width , then compare the deviation between the mean and the center point, and superimpose the total historical deviation under this dimension, and calculate it using the formula.

[0122] The sample data is shown in Table 11:

[0123] Table 11 Sample table of state values ​​for the first five cycles of cycle 5 (temperature rise dimension):

[0124] Assume that the temperature rise envelope boundary of the aggregation section is , then , the center point is , the mean shift is (1.346-1.33=0.016), and the sum of historical fluctuations is:

[0125] ;

[0126] Substituting into the formula we get:

[0127] ;

[0128] Compared with the update threshold setting value of 0.15, the update condition is met. Therefore, the recording cycle 5 is triggered once in the temperature rise dimension. If there are more than three triggers in subsequent cycles, the state envelope boundary is updated and the corrected state boundary interval is finally generated.

[0129] Dimension difference strength is used to measure the comprehensive deviation of a specific dimension in the current period relative to historical behavior and boundary benchmarks. Its essence reflects whether the dimension shows a trend of discrete or abnormal fluctuation within a certain time window. This indicator comprehensively considers two aspects of information: one is the distance between the mean value of the dimension state and the center position of the aggregation segment in the current rolling period, which represents the overall offset trend; the other is the degree of fluctuation of the historical period state value relative to the mean, reflecting local stability and disturbance activity. By weightedly combining the two parts and normalizing them to the envelope width of the aggregation segment state, the dimension difference strength can intuitively indicate whether the dimension has deviated from the original stable aggregation state and has signs of trend transfer. Therefore, the larger the value, the more likely the dimension trajectory is to be in a critical state of deviation or instability, and has a higher risk of dynamic interference.

[0130] In the formula, Represents a period The absolute deviation between the mean of the first five periods and the center of the envelope is used to characterize the deviation of the overall trajectory from the trend. It is the sum of the fluctuation amplitudes of the historical state values ​​relative to their own mean, which is used to reflect the degree of trajectory fluctuation activity. The sum of the two forms a complete periodic deviation index. The denominator is It represents the state fluctuation range of the aggregate segment in this dimension, and serves as a normalization factor to standardize the scale of the overall deviation amplitude. Therefore, the molecular offset term is scaled in fractional form to ensure comparability in different dimensions and dimensions. The pre-coefficient It is used to average the five-period historical deviations, so that the overall formula takes into account both the absolute deviation trend and the local fluctuation intensity, and uniformly adjusts the scale to construct an indicator system that reflects both the directionality of trajectory deviation and the sensitivity to local disturbances.

[0131] The fault segment marking submodule performs tracking judgment on the period that continues to deviate after the update based on the correction state boundary interval and the trajectory offset identification sequence. It detects whether the continuous offset is still in the out-of-bounds state after completing three boundary updates. If it meets the requirements, the corresponding period is marked as a fault segment period, the fault period number is output, and the centrifuge fault prediction result is obtained;

[0132] Based on the obtained corrected state boundary interval, the periodic state value is checked again to see if it still crosses the boundary after the update. The state judgment is performed for each period, and the periodic state vector is compared in each dimension to see if it exists. or situation, in which Indicates the updated boundary value. For example, the status value of cycle 6 in the temperature rise dimension is , has exceeded the updated range , is marked as out-of-bounds. If the out-of-bounds situation occurs after three boundary updates and still meets the out-of-bounds condition, the cycle will be determined as a fault cycle. The cycle number is recorded and all such cycles are summarized to finally obtain the centrifuge fault prediction result.

[0133] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A centrifuge fault prediction system based on machine learning, characterized in that: The system comprises: The data channel synchronization module obtains the continuous signal data of the starting point of the current spindle cycle of the centrifuge, divides the vibration analysis cycle into segments, and synchronizes the time scales of all signal segments to generate a unified segmented signal index matrix; The multidimensional feature extraction module calculates the time difference of the peak position of the vibration signal based on the unified segmented signal index matrix, counts the proportion of the number of mutation points in the current signal, obtains the mean of the continuous fluctuation amplitude of the speed signal, calculates the temperature rise rate index of the temperature signal, and sequentially constructs a shape-forming periodic state evolution vector sequence; The state evolution indicator construction module calculates the root mean square of the difference in each dimension as a trend change indicator based on the periodic state evolution vector sequence, records the position of the trend mutation point, and generates a trend mutation state change indicator set; The trend aggregation trajectory identification module performs cosine similarity analysis and marks aggregation segments based on the trend mutation state change indicator set, performs trajectory deviation comparison and counts the number of boundary violations, and obtains a trajectory deviation identification sequence; The fault section identification module calculates the difference based on the trajectory offset identification sequence, determines the boundary update amplitude and performs boundary replacement, marks the fault area, and generates a centrifuge fault prediction result.

2. The centrifuge fault prediction system based on machine learning according to claim 1, characterized in that: The unified segmented signal index matrix includes a periodic synchronization number, a time scale alignment identifier, a signal channel mapping relationship, a segmented time series index, and a sampling interval record; the periodic state evolution vector sequence includes a vibration position differential value, a current mutation ratio factor, a speed fluctuation mean amplitude index, a temperature rise rate change, and a time series association code; the trend mutation state change indicator set includes a trend response amplitude index, a periodic fluctuation mean square error value, an abnormal evolution position index, and an inter-vector difference expression degree; the trajectory offset identifier sequence includes a similarity aggregation label, a direction consistency count, a trajectory change interval envelope value, and an out-of-bounds period position record; the centrifuge fault prediction result includes a predicted state segment number, a fault evolution path label, a trajectory center difference metric, and a fault level discrimination result.

3. The centrifuge fault prediction system based on machine learning according to claim 1, characterized in that: The data channel synchronization module includes: The signal acquisition submodule acquires continuous signal data from the vibration acceleration sensor, axial speed recorder, current monitoring channel, and temperature sensing device. Based on the current spindle cycle starting point as a reference, it divides the vibration analysis cycle into acquisition reference points and collects all signal data segments to generate a multi-source synchronous signal segment set. The time synchronization submodule reads the timestamp data inside each signal segment based on the multi-source synchronization signal segment set, calculates the time deviation between each signal segment and the starting point of the spindle cycle, adjusts the time axis data of each signal segment based on the time reference of the starting point of the spindle cycle, compares whether the deviation exceeds the time synchronization threshold, and if so, adds or subtracts the signal time axis until the time axis of each signal segment meets the synchronization standard, thereby obtaining a unified time-aligned signal segment set; The signal index generation submodule records the start and end time points, signal length, sampling rate and channel sequence number of each signal segment according to the unified time-aligned signal segment set, combines the time points and channel numbers of each signal segment with the spindle cycle starting point reference benchmark, establishes a mapping index between the signal segment and the spindle cycle reference point, and generates a unified segmented signal index matrix.

4. The centrifuge fault prediction system based on machine learning according to claim 1, characterized in that: The multidimensional feature extraction module includes: The vibration feature calculation submodule obtains the vibration acceleration signal data segments based on the unified segmented signal index matrix, selects all vibration data points in each signal segment, detects the vibration amplitude point by point, records the time positions of all peak points, calculates the time difference between each pair of adjacent peak points, and summarizes the vibration signal peak time difference; The current and speed feature extraction submodule reads the current signal data segment based on the time difference of the vibration signal peak value, performs point-by-point amplitude difference calculation on all sampling points, determines whether the differential data exceeds the set current mutation threshold, counts the number of mutation points, and simultaneously calculates the proportion of the number of current signal mutation points. It also performs absolute value calculation on the speed data of consecutive sampling points, calculates the mean of the fluctuation amplitude, and obtains the current mutation proportion and the mean of the speed fluctuation. The temperature feature analysis submodule reads the temperature signal data segment based on the current mutation ratio and the speed fluctuation mean, performs point-by-point differentiation on the temperature values ​​of all sampling points, calculates the temperature rise rate per unit time based on the time interval, and obtains the average temperature rise rate index through averaging. Combined with the vibration signal peak time difference, the current mutation ratio and the speed fluctuation mean, the feature results are sorted in chronological order, and combined according to the time series to establish a periodic state evolution vector sequence.

5. The centrifuge fault prediction system based on machine learning according to claim 1, characterized in that: The state evolution indicator building module includes: The difference extraction submodule sequentially obtains three adjacent vector groups based on the periodic state evolution vector sequence, extracts the values ​​of each group of three vectors in the same dimension, performs a subtraction operation on the values ​​between two adjacent periods in the same dimension, repeats the operation for all dimensions, and obtains a periodic state difference sequence table; The trend calculation submodule performs a square operation and an average root operation on the difference items of each dimension according to the period state difference sequence table to obtain the root mean square trend change value under the corresponding dimension. Combined with the previous and next data of the current period, the trend change intensity is calculated to obtain the trend change intensity sequence; The mutation judgment submodule sets a fault diagnosis threshold based on the trend change intensity sequence, performs a threshold comparison operation on the trend intensity values ​​of three consecutive cycles, and judges one by one whether the three-cycle sequences are all greater than the fault diagnosis threshold. If the conditions are met, the current cycle number is recorded as the position of the trend mutation point, and the state evolution data corresponding to all mutation points are extracted and sorted to establish a trend mutation state change indicator set.

6. The centrifuge fault prediction system based on machine learning according to claim 1, characterized in that: The trend aggregation trajectory identification module includes: The directional consistency identification submodule obtains the state vectors of the two cycles before and after each mutation point based on the trend mutation state change indicator set, extracts the vector data of each dimension of the cycle, calculates the cosine value of the angle, and determines whether the cosine value is greater than or equal to the directional consistency benchmark threshold. If the condition is met, it is marked as a directional consistent dimension, otherwise it is marked as a directional inconsistent dimension, and a sequence of directional consistent dimension values ​​is obtained; The aggregation segment determination submodule obtains the direction-consistent dimension value of each cycle based on the direction-consistent dimension number value sequence, determines whether the direction-consistent dimension number of each cycle is greater than or equal to the principal component contribution rate threshold, and if the condition is met, marks the corresponding cycle location as an aggregation cycle, identifies the continuous cycle number range and sets it as an aggregation segment, compares the minimum and maximum values ​​of each dimension state value, obtains the upper and lower boundary values ​​of the state within the segment, and obtains the state envelope trajectory interval; The trajectory deviation statistics submodule judges the state vector of each subsequent cycle in turn according to the state envelope trajectory interval, compares it with the maximum and minimum values ​​of the envelope interval of the corresponding dimension, and determines whether it exceeds the interval boundary range. If the dimension state value is greater than the maximum value or less than the minimum value, the corresponding dimension is marked as an out-of-bounds dimension. The number of out-of-bounds dimensions in each cycle is counted and the out-of-bounds ratio is calculated. If it is greater than the trajectory deviation judgment reference value, the cycle state is marked as a trajectory deviation, and a trajectory deviation identification sequence is generated.

7. The centrifuge fault prediction system based on machine learning according to claim 1, characterized in that: The fault section identification module includes: The cross-border accumulation judgment submodule traverses and judges all cycles based on the trajectory offset identification sequence, reads the value of each cycle in the offset identification value sequence, and records the last cycle of the segment as the cross-border judgment trigger point if there are more than three consecutive cycles. The numbers of all cycles that meet the cross-border judgment number of more than three consecutive times are counted to obtain the trigger cycle sequence; The boundary update decision submodule reads the state vectors of the five cycles before each trigger cycle according to the trigger cycle sequence, calculates the five-cycle mean for each dimension, obtains the upper and lower boundaries of the aggregation segment state envelope trajectory interval, calculates the dimension difference strength, records the number of update marks, and uses the current mean interval as the update boundary after the cumulative number reaches three times to form a corrected state boundary interval; The fault section marking submodule performs tracking judgment on the period in which the offset continues to occur after the update based on the correction state boundary interval and the trajectory offset identification sequence. It detects whether the continuous offset is still in the out-of-bounds state after completing three boundary updates. If it meets the requirements, the corresponding period is marked as a fault section period, the fault period number is output, and the centrifuge fault prediction result is obtained.

Images