Search

CN-121412531-B - Sensor data analysis method and system based on artificial intelligence

CN121412531BCN 121412531 BCN121412531 BCN 121412531BCN-121412531-B

Abstract

The invention is suitable for the sensor data analysis field, and provides a sensor data analysis method and a sensor data analysis system based on artificial intelligence, wherein the method comprises the steps of collecting multidimensional historical sensor data of normal operation of equipment, preprocessing and extracting features to obtain a standardized historical feature matrix; the method comprises the steps of constructing a device health reference model based on the matrix, establishing multi-dimensional feature dynamic association, collecting device data in real time, extracting features according to the same flow to obtain real-time feature vectors, combining the reference model with a known fault case library, constructing a fault evolution prediction model through countermeasure training, inputting the real-time vectors to generate a future multivariable sensor data prediction sequence, calculating fault deterioration acceleration and predicting reaching a safety threshold time based on the prediction sequence, and comparing to generate a risk level signal and a maintenance strategy suggestion. The intelligent monitoring and analysis of the whole flow of the equipment are realized, and the operation and maintenance accuracy and the foresight are improved.

Inventors

  • ZHENG KAI

Assignees

  • 山东闪电云科技工程有限公司

Dates

Publication Date
20260508
Application Date
20251105

Claims (7)

  1. 1. A method for analyzing sensor data based on artificial intelligence, the method comprising: collecting multidimensional historical sensor data of monitored equipment in a normal running state, preprocessing and extracting features to obtain a standardized historical feature matrix; based on the standardized historical feature matrix, constructing an equipment health reference model, and establishing a dynamic association relation among multidimensional features through the equipment health reference model; collecting current multi-dimensional sensor data of monitored equipment in real time, and extracting the same characteristics as the multi-dimensional historical sensor data from the current multi-dimensional sensor data to obtain real-time characteristic vectors; Constructing a fault evolution prediction model based on the equipment health reference model and a known fault case library in an countermeasure training mode, inputting the real-time feature vector into the fault evolution prediction model, and generating a multivariable sensor data prediction sequence in a future designated time period; Calculating the deterioration acceleration of the fault evolution path and the time when the deterioration acceleration is expected to reach a safety threshold value based on the multivariable sensor data prediction sequence, comparing the deterioration acceleration with a preset risk level threshold value, and screening and generating a risk level signal and a maintenance strategy suggestion of the current fault evolution stage; the establishing a dynamic association relation among the multidimensional features specifically comprises the following steps: Inputting the standardized historical feature matrix into a depth self-encoder network, continuously learning low-dimensional feature representation of equipment in a normal running state, and reconstructing input features through a decoder part; Analyzing time dependence among different sensor features in a standardized historical feature matrix, capturing a multi-time-scale dependence, calculating a weight correlation matrix among vibration spectrum, temperature gradient and energy consumption efficiency features, and capturing a dynamic correlation among the vibration spectrum, the temperature gradient and the energy consumption efficiency; and establishing an equipment health reference model capable of accurately describing the multi-variable relationship under the equipment health state by combining the low-dimensional characteristic representation learned by the depth self-encoder network and the dynamic association relationship captured by the time sequence convolution network.
  2. 2. The method according to claim 1, wherein the obtaining the normalized historical feature matrix specifically comprises: collecting vibration spectrum data of equipment, temperature gradient data of equipment shell and energy consumption efficiency data of the equipment; performing time stamp alignment and outlier rejection on the acquired vibration spectrum data, temperature gradient data and energy consumption efficiency data to form historical sensor data with a unified time reference; Extracting time domain statistical features, frequency domain spectrum features and time-frequency domain wavelet energy features from historical sensor data, and constructing an initial feature set; And performing Z-score standardization processing on all the features in the initial feature set, eliminating the influence of different physical dimensions, and forming a standardized historical feature matrix.
  3. 3. The method according to claim 2, wherein the obtaining the real-time feature vector specifically comprises: synchronously acquiring real-time readings of vibration frequency spectrum, temperature gradient and energy consumption efficiency in the running of the equipment at the current moment, and adopting an outlier rejection rule which is the same as that of the historical sensor data for processing; according to the same characteristic engineering flow as the construction of the standardized historical characteristic matrix, the same characteristic extraction algorithm parameters are used for extracting time domain statistical characteristics, frequency domain spectrum characteristics and time-frequency domain wavelet energy characteristics from the processed real-time readings; and carrying out standardization processing on the extracted real-time features by using the mean and variance parameters calculated when the standardized historical feature matrix is constructed, and generating real-time feature vectors consistent with the feature space of the standardized historical feature matrix.
  4. 4. The method according to claim 3, wherein the constructing the fault evolution prediction model specifically comprises: extracting development process data of each type of fault mode from a known fault case library, and obtaining a complete evolution record of each fault type; Constructing a generator network by taking an encoder part in the equipment health reference model as an infrastructure of a discriminator network; Receiving real-time feature vectors as input through a generator network, and gradually generating a multivariable sensor data prediction sequence of a plurality of time points in the future through time step expansion; through the countermeasure training of the generator network and the discriminator network, the generator network learns to generate a fault evolution prediction model which accords with the physical rule and is consistent with the fault development mode in the known fault case library.
  5. 5. The method according to claim 4, characterized in that said calculating the deteriorated acceleration of the faulty evolution path and the time at which the safety threshold is expected to be reached, in particular comprises: Extracting performance index change curves of vibration amplitude, temperature value and energy consumption along with time from a multivariable sensor data prediction sequence; Performing second-order differential calculation on the time-varying curve of the performance index, and calculating a first-order derivative and a second-order derivative of each time point, wherein the second-order derivative represents the deteriorated acceleration of the fault development rate; Based on the time-varying curve of the performance index, fitting the development trend of the curve, determining the time point when the time-varying curve of the performance index breaks through a preset safety threshold for the first time, and calculating the time when the performance index is expected to reach the safety threshold; and (3) integrating the deteriorated acceleration of the performance index and the time when the safety threshold is expected to be reached, calculating the overall risk index, and evaluating the emergency degree and the development trend of the overall fault risk.
  6. 6. The method of claim 5, wherein the screening generates risk level signals and maintenance policy suggestions for the current fault evolution stage, specifically comprising: establishing a risk assessment matrix taking deteriorated acceleration as a vertical axis and taking time expected to reach a safety threshold as a horizontal axis, and dividing the risk assessment matrix into three areas of low risk, medium risk and high risk; Mapping the calculated deteriorated acceleration and the time expected to reach the safety threshold value into a risk assessment matrix, and determining the coordinate position of the current fault evolution stage in the risk assessment matrix; according to the risk area where the current fault evolution stage is located, converting the matrix coordinates into corresponding risk level signals through a predefined state mapping table, and generating corresponding risk level signals, including low risk signals, medium risk signals and high risk signals; Based on the generated risk level signal, matching maintenance policy records are retrieved from a relational database of a maintenance policy repository, and matching maintenance policy suggestions are screened from the maintenance policy repository.
  7. 7. An artificial intelligence based sensor data analysis system for performing the method of any of claims 1-6, the system comprising: the historical sensor data acquisition module is used for acquiring multidimensional historical sensor data of the monitored equipment in a normal running state, preprocessing and extracting features to acquire a standardized historical feature matrix; The health reference model construction module is used for constructing a device health reference model based on the standardized historical feature matrix and establishing a dynamic association relation among multidimensional features through the device health reference model; The real-time sensor data acquisition module is used for acquiring current multi-dimensional sensor data of the monitored equipment in real time, and extracting the same characteristics of the current multi-dimensional sensor data as the multi-dimensional historical sensor data to obtain a real-time characteristic vector; The fault evolution prediction module is used for constructing a fault evolution prediction model based on the equipment health reference model and a known fault case library in an countermeasure training mode, inputting the real-time feature vector into the fault evolution prediction model and generating a multivariable sensor data prediction sequence in a future designated time period; The risk assessment module is used for calculating the deterioration acceleration of the fault evolution path and the time when the deterioration acceleration is expected to reach the safety threshold value based on the multivariate sensor data prediction sequence, comparing the deterioration acceleration with a preset risk level threshold value, and screening and generating a risk level signal and a maintenance strategy suggestion of the current fault evolution stage.

Description

Sensor data analysis method and system based on artificial intelligence Technical Field The invention belongs to the field of sensor data analysis, and particularly relates to a sensor data analysis method and system based on artificial intelligence. Background Currently, continuous and stable operation demands of key equipment in the fields of industrial production, energy transmission, mechanical manufacturing and the like are increasingly improved, sensor technology is widely applied to equipment monitoring scenes because equipment operation parameters can be captured in real time, and a multi-dimensional sensor becomes a core tool for acquiring equipment operation data. Meanwhile, the application of the artificial intelligence technology in the field of data analysis is deepened continuously, the traditional data statistical analysis is gradually evolved to predictive analysis based on deep learning, more and more equipment monitoring schemes begin to try to fuse sensor data and artificial intelligence algorithms so as to realize dynamic perception of the health state of equipment and fault early warning, and the whole scheme is still in a gradual perfection stage. The prior art cannot realize fault evolution prediction from multi-dimensional sensor data standardization processing, multi-dimensional characteristic dynamic association modeling and accordance with equipment physical operation rules, to full-flow integrated analysis of risk level system judgment and targeted maintenance strategy matching, so that equipment health state depiction accuracy is insufficient, multi-dimensional collaborative state during normal operation of equipment cannot be truly reflected, fault prediction reliability is low, a dynamic process from sprouting to deterioration of faults is difficult to accurately capture, risk assessment lacks quantitative basis, only can realize judgment of whether faults exist or not, cannot define emergency degree and residual coping time of the faults, maintenance decision lacks pertinence, an adaptive operation and maintenance scheme cannot be provided according to actual risk conditions of the equipment, and requirements of key equipment refinement and prospective operation and maintenance are difficult to meet. Disclosure of Invention The invention aims to provide a sensor data analysis method based on artificial intelligence, which aims to solve the technical problems in the prior art determined in the background art. The invention is realized in such a way that an artificial intelligence based sensor data analysis method comprises the following steps: collecting multidimensional historical sensor data of monitored equipment in a normal running state, preprocessing and extracting features to obtain a standardized historical feature matrix; based on the standardized historical feature matrix, constructing an equipment health reference model, and establishing a dynamic association relation among multidimensional features through the equipment health reference model; collecting current multi-dimensional sensor data of monitored equipment in real time, and extracting the same characteristics as the multi-dimensional historical sensor data from the current multi-dimensional sensor data to obtain real-time characteristic vectors; Constructing a fault evolution prediction model based on the equipment health reference model and a known fault case library in an countermeasure training mode, inputting the real-time feature vector into the fault evolution prediction model, and generating a multivariable sensor data prediction sequence in a future designated time period; And calculating the deterioration acceleration of the fault evolution path and the time when the deterioration acceleration is expected to reach a safety threshold value based on the multivariable sensor data prediction sequence, comparing the deterioration acceleration with a preset risk level threshold value, and screening and generating a risk level signal and a maintenance strategy suggestion of the current fault evolution stage. As a further aspect of the present invention, the obtaining a normalized historical feature matrix specifically includes: collecting vibration spectrum data of equipment, temperature gradient data of equipment shell and energy consumption efficiency data of the equipment; performing time stamp alignment and outlier rejection on the acquired vibration spectrum data, temperature gradient data and energy consumption efficiency data to form historical sensor data with a unified time reference; Extracting time domain statistical features, frequency domain spectrum features and time-frequency domain wavelet energy features from historical sensor data, and constructing an initial feature set; And performing Z-score standardization processing on all the features in the initial feature set, eliminating the influence of different physical dimensions, and forming a standardized historical feature matrix. As a fu