Speed reducer fault prediction method and system based on multi-sensor fusion

Abstract

The invention provides a speed reducer fault prediction method and system based on multi-sensor fusion, and belongs to the field of fault prediction, and the method comprises the steps: S1, obtaining the operation data of a speed reducer, carrying out the preprocessing, obtaining the standardized to-be-detected data, and carrying out the hierarchical feature extraction; s2, a working condition reference library is constructed, the multi-level features are compared with the working condition reference library, the current working condition category of the speed reducer is determined, dynamic correction is carried out, and the deviation degree value of each level feature is obtained; s3, performing multi-sensor collaborative evaluation on the deviation degree value of each level feature to obtain a comprehensive fault risk score; s4, performing dynamic fault diagnosis according to the comprehensive fault risk score to obtain the fault type of the speed reducer; and S5, the fault type and the operation data of the speed reducer are stored in a fault library, and maintenance strategy suggestions are generated based on maintenance strategy self-adaption. According to the method, the faults of the speed reducer are accurately predicted and classified.

Description

Technical Field

[0001] The present invention relates to the technical field of fault prediction, and in particular to a speed reducer fault prediction method and system based on multi-sensor fusion. Background Art

[0002] Reducers are widely used in industrial production, energy, transportation and other fields. They are the core transmission components in rotating machinery. Their main function is to reduce the speed of power sources such as motors, while increasing the output torque and realizing the transformation of power characteristics. Reducers usually contain one or more gear transmission systems, which can reduce the speed and increase the torque by changing the gear diameter ratio.

[0003] As a key link in the mechanical system, the operating status of the reducer directly affects the safety and efficiency of the entire production line. Among the existing reducer fault detection methods, most use fixed thresholds, which cannot adapt to the normal parameter changes of the reducer under different loads. In addition, the reducer is prone to triggering false alarms during normal operating conditions such as start-up and shutdown processes and load fluctuations. This leads to poor adaptability of fault detection to working conditions. In addition, since existing reducer fault detection usually collects single data, such as fault identification by measuring the vibration signals of components such as gears and bearings, detecting abnormal hot spots through infrared thermal imaging or temperature sensors, and judging the name of equipment status by detecting wear particles in the oil, the signal of reducer faults is weak in the early stages, making it difficult to reliably judge with a single data signal. At the same time, different types of faults (such as minor tooth surface scratches and early bearing damage) appear similar on a single sensor, resulting in fault classification errors.

[0004] Therefore, finding a method that can not only distinguish the equipment status of the reducer under complex working conditions but also improve the accuracy of reducer fault identification is a technical problem that needs to be solved urgently by those skilled in the art. Summary of the Invention

[0005] The present invention provides a speed reducer fault prediction method and system based on multi-sensor fusion, which is used to solve the defect of relying on single data fault classification in the existing technology, realize accurate prediction and classification of speed reducer faults, and effectively solve the adaptability problem of speed reducer fault detection under different working conditions.

[0006] The present invention provides a speed reducer fault prediction method based on multi-sensor fusion, comprising the following steps: S1. Acquire operating data of a reducer through multiple sensors, preprocess the operating data of the reducer to obtain standardized data to be detected, and perform hierarchical feature extraction on the standardized data to be detected to obtain multi-level features; S2. Build a working condition benchmark library, compare the multi-level features with the working condition benchmark library respectively, determine the current working condition category of the reducer, and dynamically correct the multi-level features according to the current working condition category of the reducer to obtain the deviation value of each level feature; S3. Perform multi-sensor collaborative evaluation on the deviation values ​​of the features at each level to obtain a comprehensive fault risk score; S4. Perform dynamic fault diagnosis based on the comprehensive fault risk score to obtain the fault type of the reducer; S5. Storing the fault type and operating data of the reducer in a fault database, and generating maintenance strategy recommendations based on maintenance strategy adaptation.

[0007] According to a reducer fault prediction method based on multi-sensor fusion provided by the present invention, the multi-level features include mechanical level features, thermodynamic level features and operating condition level features, wherein: The mechanical level features include vibration energy features, vibration envelope features and spectrum features; The thermodynamic level characteristics include absolute value of temperature, rate of temperature change and temperature gradient; The operating condition level characteristics include load characteristics, speed characteristics and oil characteristics.

[0008] According to a reducer fault prediction method based on multi-sensor fusion provided by the present invention, step S2 specifically includes: Based on the multi-level features, the current operating state of the reducer is identified to obtain a current operating condition category, wherein the current operating state includes a startup transition condition, a light-load steady-state condition, a medium-load steady-state condition, a heavy-load condition, an overload condition, and a braking condition; Obtain historical operating data of the reducer, build a working condition benchmark library based on the historical operating data of the reducer, query the working condition benchmark library based on the current working condition category, obtain the normal reference range of each level feature of the current working condition category, and compare the multi-level feature values ​​with the normal reference range of each feature to obtain a working condition comparison result; Dynamic correction is used to calibrate the working condition comparison results to obtain the deviation values ​​of the characteristics at each level.

[0009] According to a reducer fault prediction method based on multi-sensor fusion provided by the present invention, the dynamic correction includes cross-influence correction, historical trend correction and seasonal environment correction, wherein: The cross-influence correction is used to adjust the correlation between the features at each level to obtain a cross-corrected feature value; The historical trend correction is used to correct the cross correction characteristic value according to the baseline formed by the long-term operation data of the reducer to obtain the trend correction characteristic value; The seasonal environmental correction is used to correct the trend correction characteristic value according to the current environmental temperature conditions to obtain the final characteristic deviation value of each level.

[0010] According to a reducer fault prediction method based on multi-sensor fusion provided by the present invention, the cross-impact correction specifically includes: Constructing a reducer multi-sensor feature association matrix based on the multi-level features, and identifying the correlation between the features of each level under the current working condition category according to the reducer multi-sensor feature association matrix; Calculating the correlation strength between each level feature and other level features based on the correlation, and determining the correction amount of each level feature; The features at each level are corrected according to the correction amount of the features at each level to obtain a cross-corrected feature value.

[0011] According to a reducer fault prediction method based on multi-sensor fusion provided by the present invention, the historical trend correction specifically includes: Obtain historical operating data of the reducer, extract an aging baseline of each level characteristic changing with operating time from the historical operating data of the reducer, and fit the aging baseline to obtain trend curve parameters; The aging characteristics in the multi-level characteristic values ​​under the current working condition category are eliminated according to the trend curve parameters to obtain the trend correction characteristic value.

[0012] According to a reducer fault prediction method based on multi-sensor fusion provided by the present invention, step S3 specifically includes: The deviation values ​​of each level feature within the reducer gear meshing cycle are synchronously detected through a sliding time window, and the correlation of the deviation values ​​of each level feature in time and space is analyzed to determine whether the features of each level originate from the same reducer fault source. Different mappings are used according to the number of reducer fault sources to obtain real-time multi-sensor collaborative scoring; Building a reducer fault mode library based on the historical operation data of the reducer, performing similarity matching between the features of each level and the reducer fault mode library, and generating a matching score for the historical fault mode of the reducer; The real-time multi-sensor collaborative score is integrated with the speed reducer historical fault pattern matching score to obtain a comprehensive fault risk score.

[0013] According to a reducer fault prediction method based on multi-sensor fusion provided by the present invention, step S4 specifically includes: Setting a risk threshold, when the comprehensive risk score exceeds the risk threshold, forming a candidate fault set based on the various hierarchical features of the reducer and the reducer fault mode library, and assigning an initial confidence level to each fault type in the candidate fault set; Continuously collect and analyze the reducer's operating data for the next time window to obtain the current reducer fault development trend; Adjusting the confidence level of the current reducer fault type based on the current reducer fault development trend and the development trend of each fault type in the candidate fault set; The fault type and severity of the current working condition category are determined based on the confidence of the current reducer fault type and the initial confidence of each fault type in the candidate fault set.

[0014] According to a method for predicting reducer faults based on multi-sensor fusion provided by the present invention, the maintenance strategy adaptation specifically includes: According to the fault type and severity of the current working condition, the remaining time for the reducer fault to develop into a critical state is evaluated and recorded as the remaining service life of the reducer; Generate differentiated maintenance recommendations for reducer failures based on the remaining service life of the reducer; Record the status of the reducer after each maintenance operation, evaluate the maintenance effect of the reducer fault, automatically adjust the evaluation standard of the fault type based on the maintenance effect of the reducer fault, and dynamically update the reducer fault maintenance recommendations according to the service life of the reducer and environmental changes.

[0015] The present invention also provides a speed reducer fault prediction system based on multi-sensor fusion, which implements the speed reducer fault prediction method as described above, including: A data processing module is used to obtain the operating data of the reducer through multiple sensors, pre-process the operating data of the reducer to obtain standardized data to be detected, and perform hierarchical feature extraction on the standardized data to be detected to obtain multi-level features; A dynamic correction module is used to build a working condition benchmark library, compare the multi-level features with the working condition benchmark library respectively, determine the current working condition category of the reducer, and dynamically correct the multi-level features according to the current working condition category of the reducer to obtain the deviation value of the features at each level; A risk assessment module is used to perform multi-sensor collaborative amplification on the deviation values ​​of the features at each level to obtain a comprehensive fault risk score; A fault classification module is used to perform dynamic fault diagnosis based on the comprehensive fault risk score to obtain the fault type of the reducer; The strategy generation module is used to store the fault type and operation data of the reducer in a fault library, and generate maintenance strategy recommendations based on maintenance strategy adaptation.

[0016] The present invention provides a reducer fault prediction method and system based on multi-sensor fusion, which realizes accurate prediction and classification of reducer faults by dynamically correcting and collaboratively evaluating the multi-dimensional characteristics of the mechanical level, thermodynamic level and operating condition level. The use of a working condition benchmark library and dynamic correction effectively solves the adaptability problem of reducer fault detection under different working conditions; at the same time, through the spatiotemporal correlation analysis of multi-sensor information, the sensitivity and accuracy of early fault identification of reducers are significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 It is a flow chart of a reducer fault prediction method based on multi-sensor fusion provided by the present invention. DETAILED DESCRIPTION

[0019] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0020] like Figure 1 As shown, the present invention provides a speed reducer fault prediction method based on multi-sensor fusion, comprising the following steps: S1. Acquire the operating data of the reducer through multiple sensors, pre-process the operating data of the reducer to obtain standardized data to be detected, and perform hierarchical feature extraction on the standardized data to be detected to obtain multi-level features.

[0021] Among them, the multiple sensors include vibration sensors, temperature sensors, torque sensors and acoustic emission sensors. That is, vibration sensors are arranged on the reducer bearing seat and gearbox housing to capture gear meshing vibration and bearing rolling abnormalities. Temperature sensors are set in the lubricating oil pipeline or bearing cavity to measure the temperature in real time and reflect the thermal state. Torque sensors are installed on the input shaft or output shaft to identify large fluctuations in load conditions (such as starting impact, overload, etc.), and acoustic emission transmitters are installed to collect the impact sound signals of micro-cracks in the reducer.

[0022] Understandably, while reducer bearing faults and gear faults may share some similar vibration characteristics, they differ significantly in temperature and load response. Early fault signals are often weak and can be masked by a single-level feature. Therefore, the present invention, through hierarchical feature extraction of reducer operating data, is able to distinguish between fault types with similar manifestations but different mechanisms, and amplify weak fault signals. For example, a tiny gear crack may not be apparent in vibration signatures, but a collaborative analysis combining acoustic emission signatures with load fluctuation signatures can significantly improve detection sensitivity.

[0023] In one embodiment of the present invention, preprocessing includes digital filtering, digital normalization and signal segmentation. Bandpass filtering is used to remove high-frequency noise and low-frequency interference signals, adjust different sensor data to a unified range, and divide data segments according to the sampling frequency and the reducer working cycle to improve data processing efficiency.

[0024] Specifically, the multi-level features include mechanical level features, thermodynamic level features and operating condition level features, wherein: The mechanical level features include vibration energy features, vibration envelope features and spectrum features; The thermodynamic level characteristics include absolute value of temperature, rate of temperature change and temperature gradient; The operating condition level characteristics include load characteristics, speed characteristics and oil characteristics.

[0025] It can be understood that the mechanical level mainly processes vibration data and acoustic emission data, and the mechanical level characteristics can directly reflect the physical quantities of the mechanical state of the reducer. The thermodynamic level mainly processes temperature data, and the thermodynamic level characteristics can reflect the indicators of the energy conversion and thermal equilibrium state of the reducer. The operating condition level mainly processes torque data and operating parameters, and the operating condition level characteristics reflect the indicators of the overall working state of the reducer.

[0026] This method extracts features from reducer operating data at the mechanical, thermodynamic, and operating condition levels, achieving a multi-dimensional understanding of the reducer's status. Compared to single-feature analysis, multi-level features can comprehensively reflect the state changes of each reducer component, improving fault coverage, especially for complex faults involving multiple components.

[0027] In one embodiment of the present invention, the extraction of mechanical level features includes vibration energy feature extraction, vibration envelope feature extraction and spectrum feature extraction, wherein the vibration energy feature extraction specifically includes: calculating time domain statistical parameters, including root mean square value, peak value, peak-to-peak value, peak factor, pulse factor and margin factor, dividing the vibration signal into multiple frequency bands (low frequency 0-500Hz, medium frequency 500-2000Hz, high frequency 2000-5000Hz), calculating the energy proportion of each frequency band, and for the acoustic emission signal, calculating the acoustic energy accumulation value and counting rate for early fatigue crack detection; the vibration envelope feature extraction specifically includes: using Hilbert transform The vibration signal of the reducer is subjected to envelope analysis to extract the modulation signal features, and the amplitudes of the characteristic frequencies of the bearing inner / outer ring, the cage characteristic frequency and the gear meshing characteristic frequency in the envelope spectrum are calculated. The envelope modulation depth and modulation index are extracted to evaluate the state of the rolling elements of the reducer bearing; the spectrum feature extraction specifically includes: performing fast Fourier transform on the vibration signal of the reducer, calculating the gear meshing frequency and its harmonic amplitude, extracting the ratio of the side band energy of the reducer gear meshing frequency to the center frequency energy, calculating the spectrum kurtosis and spectrum entropy, quantifying the spectrum shape features, and performing inverse spectrum analysis on the spectrum to extract the periodic faults of the reducer gear.

[0028] In one embodiment of the present invention, the extraction of thermodynamic hierarchical features includes temperature absolute value feature extraction, temperature change rate feature extraction and temperature gradient feature extraction, wherein the temperature absolute value feature extraction specifically includes: extracting the average temperature and maximum temperature of each measuring point of the reducer (input bearing, output bearing, gear box, lubricating oil), calculating the difference between the temperature of each measuring point and the ambient temperature, and extracting the duration ratio of the temperature exceeding the preset threshold; the temperature change rate feature extraction specifically includes: calculating the change rate of the temperature of each measuring point of the reducer within a short time window (10 minutes), extracting the maximum value, average value and standard deviation of the temperature change rate, identifying abnormal heating trends, and calculating the fluctuation coefficient of the temperature change rate; the temperature gradient feature extraction specifically includes: calculating the temperature difference between the input / output bearings of the reducer and the temperature difference between different positions of the gear box, constructing a temperature gradient vector, and extracting the temperature gradient change trend.

[0029] In one embodiment of the present invention, the operating condition level feature extraction includes load feature extraction, speed feature extraction and oil feature extraction, wherein the load feature extraction specifically includes: calculating the average torque, peak torque and torque fluctuation coefficient of the reducer, extracting the load change rate and load cycle characteristics (frequency, amplitude), and calculating the input-output power ratio of the reducer; the speed feature extraction specifically includes: extracting the average speed of the reducer, the speed fluctuation amplitude and the speed change pattern, calculating the acceleration and deceleration values, identifying the start-stop characteristics of the reducer, and extracting the response relationship between the speed change and the vibration and temperature change; the oil feature extraction specifically includes: extracting the oil temperature, pressure and flow parameters according to the oil sensor data, calculating the oil viscosity change index, and extracting the oil contamination index to evaluate the lubrication quality.

[0030] S2. Build a working condition reference library, compare the multi-level features with the working condition reference library respectively, determine the current working condition category of the reducer, and dynamically correct the multi-level features according to the current working condition category of the reducer to obtain the deviation value of each level feature.

[0031] Furthermore, step S2 specifically includes: Based on the multi-level features, the current operating state of the reducer is identified to obtain a current operating condition category, wherein the current operating state includes a startup transition condition, a light-load steady-state condition, a medium-load steady-state condition, a heavy-load condition, an overload condition, and a braking condition; Obtain historical operating data of the reducer, build a working condition benchmark library based on the historical operating data of the reducer, query the working condition benchmark library based on the current working condition category, obtain the normal reference range of each level feature of the current working condition category, and compare the multi-level feature values ​​with the normal reference range of each feature to obtain a working condition comparison result; Dynamic correction is used to calibrate the working condition comparison results to obtain the deviation values ​​of the characteristics at each level.

[0032] The calculation formula of the deviation value is: in, Represents the initial deviation value of the i-th level feature, represents the measured value of the i-th level feature, represents the baseline mean of the i-th level feature under the current working condition category, Represents the benchmark standard deviation of the i-th level feature under the current working condition category.

[0033] It can be understood that the benchmark mean and benchmark standard deviation can be determined based on historical operating data statistics.

[0034] In one embodiment of the present invention, the operating condition category is determined using speed and load characteristics within the operating condition hierarchy. Specifically, the speed time series curve slope is analyzed to identify the reducer's operating phase. A sustained rapid increase in speed and large torque fluctuations are detected as a "startup transition condition." A sustained decrease in speed accompanied by energy feedback characteristics (such as a brief negative input shaft torque or direction reversal, a change in current direction, or abnormal voltage parameters) is identified as a "braking condition." Relatively stable speed phases are classified as steady-state conditions, further subdivided based on load factor.

[0035] For the subdivision of steady-state operating conditions, the load factor is calculated as the ratio of the current average torque to the rated torque of the reducer. When the load factor does not exceed 30%, the current operating condition is identified as a "light-load steady-state condition." When the load factor is between 30% and 70%, it is identified as a "medium-load steady-state condition." When the load factor is between 70% and 100%, it is identified as a "heavy-load condition." When the load factor exceeds 100%, it is identified as an "overload condition." In practice, the operating condition identification results are corrected based on characteristics such as the torque fluctuation coefficient and oil temperature to improve the accuracy of operating condition identification. For example, if the reducer's speed characteristics at a certain moment show a stable input shaft speed of around 1450 rpm, a torque fluctuation coefficient of 0.15, and an average torque of 64% of the rated torque, the operating condition is identified as a "medium-load steady-state condition." If the reducer's speed at another moment shows a significant downward trend accompanied by a brief period of negative torque, the condition is identified as a "braking condition."

[0036] It is understandable that the specific data basis for working condition identification and the normal reference range of each level feature in the working condition reference library can be set according to the actual use of the reducer, and the present invention does not impose specific limitations on this.

[0037] The working condition identification is described in detail using a specific embodiment, where the gear ratio of the reducer is 4.5, the rated torque is 5000 N·m, and the rated input speed is 1480 rpm: Extract multi-level features of the reducer, including: average speed of 1450 rpm, average torque of 3200 N·m, torque fluctuation coefficient of 0.15, bearing RMS vibration value of 1.8 mm / s, gear meshing frequency amplitude of 0.9 g, input bearing temperature of 58°C, and ambient temperature of 25°C; Based on the speed and torque characteristics, the reducer is judged to be in the steady-state operation stage, and the load factor is calculated to be 64% (3200 / 5000×100%), which is a medium-load steady-state operating condition; Query the operating condition benchmark database to obtain the normal reference ranges for each characteristic under medium-load steady-state conditions: bearing RMS vibration value range [0.7, 2.1] mm / s (mean 1.4, standard deviation 0.23), gear mesh frequency amplitude range [0.4, 1.2] g (mean 0.8, standard deviation 0.13), input bearing temperature range [45, 65] °C (mean 55, standard deviation 3.33); Calculate the initial deviation of each level feature: the deviation of the bearing RMS vibration value is 0.58, the deviation of the gear meshing frequency amplitude is 0.26, and the deviation of the input bearing temperature is 0.30; Dynamic corrections were made to the initial deviations of the features at each level: Considering that the ambient temperature was 25°C, 20°C higher than the baseline ambient temperature, the input bearing temperature deviation was corrected to -0.10; considering that the equipment had been in operation for 18 months, the baseline mean of the bearing RMS vibration value was corrected to 1.48 mm / s, and the corresponding deviation was corrected to 0.46; considering that the torque fluctuation coefficient was 0.15, the gear mesh frequency amplitude deviation was corrected to 0.23; Output the corrected characteristic deviation values ​​of each level: bearing RMS vibration value deviation is 0.46, gear meshing frequency amplitude deviation is 0.23, and input bearing temperature deviation is -0.10.

[0038] In one embodiment of the present invention, the dynamic correction includes cross-impact correction, historical trend correction, and seasonal environment correction, wherein: The cross-influence correction is used to adjust the correlation between the features at each level to obtain a cross-corrected feature value; The historical trend correction is used to correct the cross correction characteristic value according to the baseline formed by the long-term operation data of the reducer to obtain the trend correction characteristic value; The seasonal environmental correction is used to correct the trend correction characteristic value according to the current environmental temperature conditions to obtain the final characteristic deviation value of each level.

[0039] The present invention can dynamically adjust the evaluation criteria of characteristic deviation through cross-influence correction, historical trend correction and seasonal environment correction, effectively eliminating the interference of working condition changes, equipment aging and ambient temperature on reducer fault judgment.

[0040] Furthermore, the cross-impact correction specifically includes: Based on the multi-level features, a reducer multi-sensor feature association matrix is ​​constructed, and the correlation between the features of each level under the current working condition category is identified according to the reducer multi-sensor feature association matrix; the calculation formula is: in, represents the Pearson correlation coefficient between level feature i and level feature j, represents the value of level feature i in the k-th sample, represents the mean of level feature i, represents the total number of samples, represents the value of level feature j in the k-th sample, represents the mean of level feature j; Based on the correlation, the correlation strength between each level feature and other level features is calculated to determine the correction amount of each level feature; the calculation formula is: in, represents the cross-influence correction of level feature i, represents the impact factor, Indicates the current deviation value of level feature j, represents the reference deviation value of level feature j; The features at each level are corrected according to the correction amount of the features at each level to obtain the cross-corrected feature value. The calculation formula is: in, represents the cross-corrected eigenvalue of level feature i, Represents the initial deviation value of level feature i.

[0041] Understandably, when When it is greater than 0.5, cross-influence correction is performed. It can be set according to the actual situation, and different working conditions require adjustment. It is set based on the baseline value under normal conditions.

[0042] In one embodiment of the present invention, the method for constructing a feature correlation matrix is ​​as follows: for each operating condition category, the Pearson correlation coefficients between different features are calculated to form an n×n correlation matrix R (n is the total number of features), where the matrix elements are Represents the correlation coefficient between level feature i and level feature j, if If the value of the value of the level feature i is greater than the preset threshold, it is considered that there is a significant correlation between the level feature i and the level feature j. The preset threshold can be set according to actual usage requirements.

[0043] Furthermore, the historical trend correction specifically includes: Obtain the historical operating data of the reducer, and extract the aging baseline of each level feature changing with the operating time from the historical operating data of the reducer, and fit the aging baseline to obtain the trend curve parameters; the trend curve is ,in, represents the aging trend function, t represents the running time, a represents the trend intensity coefficient, b represents the time index, and c represents the initial offset; According to the trend curve parameters, the aging characteristics in the multi-level characteristic values ​​under the current working condition category are eliminated to obtain the trend correction characteristic value. The calculation formula is: in, represents the corrected baseline mean of level feature i at time t, represents the initial baseline mean of level feature i, represents the aging trend function value of level feature i, represents the trend correction characteristic value of level feature i, represents the measured value of level feature i, Represents the baseline standard deviation of level feature i.

[0044] It can be understood that the trend intensity coefficient a, the time index b and the initial offset c can be obtained by fitting based on historical operating data.

[0045] In one embodiment of the present invention, the method for extracting the aging baseline is as follows: historical data of the reducer during normal operation is arranged in chronological order, and the characteristic average value of each month (or other appropriate time period) is calculated to form a feature-time series; these sequences are fitted using an exponential or polynomial function to obtain a trend curve describing the change of characteristics over time.

[0046] In one embodiment of the present invention, seasonal environment correction specifically includes: Record the current ambient temperature in real time and determine the baseline ambient temperature based on the historical operating data of the reducer; Analyze the sensitivity of ambient temperature to different hierarchical features and establish a linear relationship between hierarchical features and ambient temperature; According to the linear relationship, the features of different levels are corrected to obtain the final deviation value of the features of each level. The calculation formula is: in, represents the sensitivity coefficient of feature i to ambient temperature, represents the deviation value of level feature i, represents the ambient temperature variable, represents the ambient temperature correction value of level feature i, Indicates the current ambient temperature. Indicates the reference ambient temperature, Indicates the final corrected deviation value, represents the trend correction characteristic value of level feature i, Indicates the ambient temperature correction amount.

[0047] Understandably, the reference ambient temperature Determined based on historical operating data, usually 20°C. Sensitivity coefficient Set according to actual usage.

[0048] S3. Perform multi-sensor collaborative evaluation on the deviation values ​​of the features at each level to obtain a comprehensive fault risk score.

[0049] Specifically, step S3 includes: The deviation values ​​of each level feature within the reducer gear meshing cycle are synchronously detected through a sliding time window, and the correlation of the deviation values ​​of each level feature in time and space is analyzed to determine whether the features of each level originate from the same reducer fault source. Different mappings are used according to the number of reducer fault sources to obtain real-time multi-sensor collaborative scoring; Building a reducer fault mode library based on the historical operation data of the reducer, performing similarity matching between the features of each level and the reducer fault mode library, and generating a matching score for the historical fault mode of the reducer; The real-time multi-sensor collaborative score is integrated with the speed reducer historical fault pattern matching score to obtain a comprehensive fault risk score.

[0050] As you can understand, each fault mode in the reducer fault pattern library contains the deviation value distribution, time evolution patterns, and interrelationships of multiple levels of features at different stages of fault development. For example, for bearing faults, the pattern library records typical patterns of vibration characteristics, temperature changes, lubrication status, and efficiency degradation at each stage, from early microcracks to severe damage.

[0051] This invention achieves collaborative amplification of weak fault signals by analyzing the temporal and spatial correlations of features at different levels. In the early stages of a fault (such as a microcrack in a gear), a single sensor signal may not reach the alarm threshold. However, through multi-sensor collaborative amplification, potential faults can be detected early, providing ample lead time for preventative maintenance.

[0052] In one embodiment of the present invention, when the reducer fault source is a single fault source, a weighted summation mapping is used. For example, since bearing faults typically cause characteristic frequency vibration, temperature rise, and lubrication state changes, the vibration feature is given a higher weight (e.g., 0.5), the temperature feature is given a medium weight (e.g., 0.3), and the operating condition feature is given a lower weight (e.g., 0.2). The deviation values ​​of the features at each level are multiplied by the response weight and then summed to obtain the collaborative score for the single fault source.

[0053] In one embodiment of the present invention, when there are multiple sources of reducer faults, a nonlinear mapping method is adopted, such as the maximum priority method, that is, the score of each fault source is calculated separately, and then the maximum value is taken as the main score, and a mutual influence factor is added to reflect the increased risk brought about by the coexistence of multiple faults.

[0054] In one embodiment of the present invention, the present invention performs similarity matching between features of each level and the reducer fault mode library, wherein the similarity calculation method can be based on Euclidean distance, Mahalanobis distance or cosine similarity, etc., and the present invention does not impose specific limitations on this.

[0055] In one embodiment of the present invention, the real-time multi-sensor collaborative score is fused with the reducer historical fault pattern matching score, and the fusion adopts adaptive weight fusion, that is, the weights of the two scores are dynamically adjusted according to the similarity between the quality of the real-time multi-sensor data and the historical fault pattern of the reducer. For example, when the real-time multi-sensor data has a high signal-to-noise ratio and the sensor is in good working condition, the weight of the real-time multi-sensor collaborative score is increased; when the current working condition has a high similarity with the reducer fault pattern library, the weight of the reducer historical fault pattern matching score is increased.

[0056] S4. Perform dynamic fault diagnosis based on the comprehensive fault risk score to obtain the fault type of the reducer.

[0057] Specifically, step S4 includes: Setting a risk threshold, when the comprehensive risk score exceeds the risk threshold, forming a candidate fault set based on the various hierarchical features of the reducer and the reducer fault mode library, and assigning an initial confidence level to each fault type in the candidate fault set; Continuously collect and analyze the reducer's operating data for the next time window to obtain the current reducer fault development trend; Adjusting the confidence level of the current reducer fault type based on the current reducer fault development trend and the development trend of each fault type in the candidate fault set; The fault type and severity of the current working condition category are determined based on the confidence of the current reducer fault type and the initial confidence assigned to each fault type in the candidate fault set.

[0058] The next time window may be 24 hours or one week, and the present invention does not impose any specific limitation on this.

[0059] The present invention adopts a dynamic fault diagnosis method of continuously collecting and analyzing reducer data, which can dynamically evaluate the fault development trend and adjust the confidence level of the fault type, thereby improving the accuracy of fault type identification.

[0060] As you can understand, the initial confidence level is determined based on the degree of match between the current hierarchical features and the typical features in the reducer fault pattern library. This invention uses a fuzzy matching algorithm to calculate the similarity between the current feature pattern and each fault pattern in the reducer fault pattern library and converts this similarity into an initial confidence level. The confidence level ranges from 0 to 100%, indicating the degree of certainty in the fault type. For example, if a bearing outer race fault is detected and the feature match reaches 70%, the initial confidence level might be set to 65%. If an abnormality in the gear mesh frequency is also observed, "gear wear" is selected as a candidate fault, but because the match is only 40%, the initial confidence level might be only 35%.

[0061] In one embodiment of the present invention, the current speed reducer fault development trend primarily includes the rate of change, direction of change, and pattern of change of characteristic deviations. By calculating the first-order derivative (rate of change) and second-order derivative (acceleration of change) of the characteristic deviations, the speed of fault development and the acceleration / deceleration trend are determined. Furthermore, the temporal evolution of the relationships between features at different levels is analyzed, such as the temporal correlation between mechanical vibration and temperature characteristics.

[0062] In one embodiment of the present invention, the fault type with the highest confidence level in the candidate fault set and exceeding a specific threshold is determined to be the primary fault type. If the confidence levels of multiple fault types exceed the threshold, the fault is determined to be a composite fault. Fault severity is categorized into four levels: minor (confidence level >75%, feature deviation <0.7, and slow growth), moderate (confidence level >80%, feature deviation 0.7-0.85, or moderate growth), severe (confidence level >85%, feature deviation >0.85, or rapid growth), and critical (confidence level >90%, feature deviation close to 1, and very rapid growth).

[0063] S5. Storing the fault type and operating data of the reducer in a fault database, and generating maintenance strategy recommendations based on maintenance strategy adaptation.

[0064] The maintenance strategy adaptation specifically includes: According to the fault type and severity of the current working condition, the remaining time for the reducer fault to develop into a critical state is evaluated and recorded as the remaining service life of the reducer; Generate differentiated maintenance recommendations for reducer failures based on the remaining service life of the reducer: It is recommended to increase the monitoring frequency and carry out preventive planning for early failure of reducers; For mid-term failures of the reducer, it is recommended to shut down the reducer for maintenance in the next time window; For late-stage failures of the reducer, it is recommended to stop the machine for inspection and repair immediately; Record the status of the reducer after each maintenance operation, evaluate the maintenance effect of the reducer fault, automatically adjust the evaluation standard of the fault type based on the maintenance effect of the reducer fault, and dynamically update the reducer fault maintenance recommendations according to the service life of the reducer and environmental changes.

[0065] Differentiated maintenance recommendations are automatically generated based on the fault type and severity, and fault assessment criteria are dynamically adjusted based on maintenance results. Based on adaptive maintenance strategies, a balanced optimization of maintenance costs and equipment reliability is achieved.

[0066] Understandably, assessing the remaining time for a fault to develop to a critical state requires consideration of multiple factors, including: the current operating condition category (fault development rates vary under different operating conditions), environmental conditions (such as temperature, humidity, etc.), the age of the reducer (older reducers generally develop faults faster), and historical maintenance (components that frequently fail may have systemic problems).

[0067] The present invention realizes accurate prediction and classification of reducer faults by dynamically correcting and collaboratively analyzing the multi-dimensional characteristics of the mechanical level, thermodynamic level and operating condition level. The use of an operating condition benchmark library and dynamic correction effectively solves the adaptability problem of reducer fault detection under different operating conditions. At the same time, through the spatiotemporal correlation analysis of multi-sensor information, the sensitivity and accuracy of early fault identification of the reducer are significantly improved, providing a reliable basis for reducer maintenance decisions, reducing the probability of unplanned equipment shutdown, extending the service life of the reducer, and reducing maintenance costs.

[0068] The present invention also provides a speed reducer fault prediction system based on multi-sensor fusion, which implements the speed reducer fault prediction method as described above, including: A data processing module is used to obtain the operating data of the reducer through multiple sensors, pre-process the operating data of the reducer to obtain standardized data to be detected, and perform hierarchical feature extraction on the standardized data to be detected to obtain multi-level features; A dynamic correction module is used to build a working condition benchmark library, compare the multi-level features with the working condition benchmark library respectively, determine the current working condition category of the reducer, and dynamically correct the multi-level features according to the current working condition category of the reducer to obtain the deviation value of the features at each level; A risk assessment module is used to perform multi-sensor collaborative amplification on the deviation values ​​of the features at each level to obtain a comprehensive fault risk score; A fault classification module is used to perform dynamic fault diagnosis based on the comprehensive fault risk score to obtain the fault type of the reducer; The strategy generation module is used to store the fault type and operation data of the reducer in a fault library, and generate maintenance strategy recommendations based on maintenance strategy adaptation.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A speed reducer fault prediction method based on multi-sensor fusion, characterized in that: The following steps are involved: S1. Acquire operating data of a reducer through multiple sensors, preprocess the operating data of the reducer to obtain standardized data to be detected, and perform hierarchical feature extraction on the standardized data to be detected to obtain multi-level features; S2. Build a working condition benchmark library, compare the multi-level features with the working condition benchmark library respectively, determine the current working condition category of the reducer, and dynamically correct the multi-level features according to the current working condition category of the reducer to obtain the deviation value of each level feature; S3. Perform multi-sensor collaborative evaluation on the deviation values ​​of the features at each level to obtain a comprehensive fault risk score; S4. Perform dynamic fault diagnosis based on the comprehensive fault risk score to obtain the fault type of the reducer; S5. Storing the fault type and operating data of the reducer in a fault database, and generating maintenance strategy recommendations based on maintenance strategy adaptation.

2. A speed reducer fault prediction method based on multi-sensor fusion according to claim 1, characterized in that: The multi-level features include mechanical level features, thermodynamic level features and operating condition level features, wherein: The mechanical level features include vibration energy features, vibration envelope features and spectrum features; The thermodynamic level characteristics include absolute value of temperature, rate of temperature change and temperature gradient; The operating condition level characteristics include load characteristics, speed characteristics and oil characteristics.

3. The method for predicting reducer faults based on multi-sensor fusion according to claim 1, characterized in that: Step S2 specifically includes: Based on the multi-level features, the current operating state of the reducer is identified to obtain a current operating condition category, wherein the current operating state includes a startup transition condition, a light-load steady-state condition, a medium-load steady-state condition, a heavy-load condition, an overload condition, and a braking condition; Obtain historical operating data of the reducer, build a working condition benchmark library based on the historical operating data of the reducer, query the working condition benchmark library based on the current working condition category, obtain the normal reference range of each level feature of the current working condition category, and compare the multi-level feature values ​​with the normal reference range of each feature to obtain a working condition comparison result; Dynamic correction is used to calibrate the working condition comparison results to obtain the deviation values ​​of the characteristics at each level.

4. A speed reducer fault prediction method based on multi-sensor fusion according to claim 3, characterized in that: The dynamic correction includes cross-influence correction, historical trend correction and seasonal environment correction, among which, The cross-influence correction is used to adjust the correlation between the features at each level to obtain a cross-corrected feature value; The historical trend correction is used to correct the cross correction characteristic value according to the baseline formed by the long-term operation data of the reducer to obtain the trend correction characteristic value; The seasonal environmental correction is used to correct the trend correction characteristic value according to the current environmental temperature conditions to obtain the final characteristic deviation value of each level.

5. The method for predicting reducer faults based on multi-sensor fusion according to claim 4, characterized in that: The cross-impact correction specifically includes: Constructing a reducer multi-sensor feature association matrix based on the multi-level features, and identifying the correlation between the features of each level under the current working condition category according to the reducer multi-sensor feature association matrix; Calculating the correlation strength between each level feature and other level features based on the correlation, and determining the correction amount of each level feature; The features at each level are corrected according to the correction amount of the features at each level to obtain a cross-corrected feature value.

6. A speed reducer fault prediction method based on multi-sensor fusion according to claim 5, characterized in that: The historical trend correction specifically includes: Obtain historical operating data of the reducer, extract an aging baseline of each level characteristic changing with operating time from the historical operating data of the reducer, and fit the aging baseline to obtain trend curve parameters; The aging characteristics in the multi-level characteristic values ​​under the current working condition category are eliminated according to the trend curve parameters to obtain the trend correction characteristic value.

7. The method for predicting reducer faults based on multi-sensor fusion according to claim 3, characterized in that: Step S3 specifically includes: The deviation values ​​of each level feature within the reducer gear meshing cycle are synchronously detected through a sliding time window, and the correlation of the deviation values ​​of each level feature in time and space is analyzed to determine whether the features of each level originate from the same reducer fault source. Different mappings are used according to the number of reducer fault sources to obtain real-time multi-sensor collaborative scoring; Building a reducer fault mode library based on the historical operation data of the reducer, performing similarity matching between the features of each level and the reducer fault mode library, and generating a matching score for the historical fault mode of the reducer; The real-time multi-sensor collaborative score is integrated with the speed reducer historical fault pattern matching score to obtain a comprehensive fault risk score.

8. The method for predicting reducer faults based on multi-sensor fusion according to claim 7, characterized in that: Step S4 specifically includes: Setting a risk threshold, when the comprehensive risk score exceeds the risk threshold, forming a candidate fault set based on the various hierarchical features of the reducer and the reducer fault mode library, and assigning an initial confidence level to each fault type in the candidate fault set; Continuously collect and analyze the reducer's operating data in the next time window to obtain the current reducer fault development trend; Adjusting the confidence level of the current reducer fault type based on the current reducer fault development trend and the development trend of each fault type in the candidate fault set; The fault type and severity of the current working condition category are determined based on the confidence of the current reducer fault type and the initial confidence of each fault type in the candidate fault set.

9. The method for predicting reducer faults based on multi-sensor fusion according to claim 1, characterized in that: The maintenance strategy adaptation specifically includes: According to the fault type and severity of the current working condition, the remaining time for the reducer fault to develop into a critical state is evaluated and recorded as the remaining service life of the reducer; Generate differentiated maintenance recommendations for reducer failures based on the remaining service life of the reducer; Record the status of the reducer after each maintenance operation, evaluate the maintenance effect of the reducer fault, automatically adjust the evaluation standard of the fault type based on the maintenance effect of the reducer fault, and dynamically update the reducer fault maintenance recommendations according to the service life of the reducer and environmental changes.

10. A speed reducer fault prediction system based on multi-sensor fusion, characterized in that: Implementing the speed reducer fault prediction method according to any one of claims 1 to 9 comprises: A data processing module is used to obtain the operating data of the reducer through multiple sensors, pre-process the operating data of the reducer to obtain standardized data to be detected, and perform hierarchical feature extraction on the standardized data to be detected to obtain multi-level features; A dynamic correction module is used to build a working condition benchmark library, compare the multi-level features with the working condition benchmark library respectively, determine the current working condition category of the reducer, and dynamically correct the multi-level features according to the current working condition category of the reducer to obtain the deviation value of each level feature; A risk assessment module is used to perform multi-sensor collaborative amplification on the deviation values ​​of the features at each level to obtain a comprehensive fault risk score; A fault classification module is used to perform dynamic fault diagnosis based on the comprehensive fault risk score to obtain the fault type of the reducer; The strategy generation module is used to store the fault type and operation data of the reducer in a fault library, and generate maintenance strategy recommendations based on maintenance strategy adaptation.

Images