Search

CN-122020167-A - Industrial equipment health state prediction method based on multi-source data fusion

CN122020167ACN 122020167 ACN122020167 ACN 122020167ACN-122020167-A

Abstract

The invention relates to the field of health monitoring, and discloses an industrial equipment health state prediction method based on multi-source data fusion, which comprises the steps of firstly synchronously collecting electric performance monitoring data and insulation state monitoring data of a transformer; fusing the characteristics of the electrical performance monitoring data and the insulation state monitoring data, inputting the characteristics into a pre-trained health evaluation model, and calculating to obtain a comprehensive health index; judging whether the equipment enters an abnormal state according to a comparison result of the comprehensive health index and a preset threshold value, then continuously calculating a short-term change trend of the electrical performance monitoring data when the industrial equipment does not enter the abnormal state, dynamically adjusting the acquisition frequency or the scanning intensity of the insulation state monitoring data according to the numerical value of the short-term change trend, and automatically focusing resources to conduct deep exploration when the abnormal trend is detected, and canceling high-density scanning when the equipment state is stable so as to improve the utilization rate of calculation resources and communication bandwidth.

Inventors

  • LIU QIANG
  • XU JIAYAN
  • WU KUI

Assignees

  • 内蒙古华桉信息科技有限公司

Dates

Publication Date
20260512
Application Date
20260127

Claims (9)

  1. 1. The industrial equipment health state prediction method based on multi-source data fusion is characterized by comprising the following steps of: S1, synchronously acquiring electrical performance monitoring data and insulation state monitoring data of a transformer; s2, fusing the characteristics of the electrical performance monitoring data and the insulation state monitoring data, inputting the characteristics into a pre-trained health evaluation model, and calculating to obtain a comprehensive health index; s3, continuously calculating short-term change trend of the electrical performance monitoring data when the industrial equipment does not enter an abnormal state, and dynamically adjusting acquisition frequency or scanning intensity of the insulation state monitoring data according to the numerical value of the short-term change trend; Wherein, the numerical value of the short-term variation trend is in negative correlation with the acquisition frequency or the scanning intensity.
  2. 2. The method for predicting the health status of industrial equipment based on multi-source data fusion according to claim 1, wherein the specific step of dynamically adjusting the collection frequency or the scanning intensity of the insulation status monitoring data in step S3 is as follows: When the value of the short-term variation trend is smaller than a first trend threshold value, the sampling frequency of the online oil chromatograph or the detection sensitivity and the scanning density of the partial discharge sensor are improved; And when the value of the short-term variation trend is larger than a second trend threshold value, reducing the sampling frequency of the online oil chromatograph or the detection sensitivity and the scanning density of the partial discharge sensor.
  3. 3. The method for predicting the health of an industrial device based on multi-source data fusion of claim 2, further comprising: S4, when the comprehensive health index exceeds a preset alarm threshold, positioning is conducted on the inside of the transformer oil tank, wherein the positioning steps are as follows: s41, performing kth scanning on the target area, dividing grids, calculating local abnormality indexes of the grids, and identifying kth hot spot units; s42, re-defining a reduced scanning area by taking the kth hot spot unit as a center; S43, executing a (k+1) th scanning in the reduced scanning area, wherein the resolution of the (k+1) th scanning is higher than that of the (k) th scanning, calculating local abnormality indexes of grids and identifying a (k+1) th hot spot unit; and S44, making k=k+1, iteratively executing the steps S42 and S43 until the physical size of the scanning area is smaller than or equal to a preset precision threshold value, and outputting the central coordinate of the hot spot unit identified by the last iteration as the final abnormal part coordinate.
  4. 4. The method for predicting the health of an industrial device based on multi-source data fusion according to claim 3, wherein in the step S43, when the (k+1) th scan is performed, the method further comprises the steps of: Dividing the current scanning area into grid levels with low, medium and high resolutions; Respectively calculating and identifying hot spot units under each level; if the high resolution level hot spot unit is included in the low resolution level hot spot unit range, determining that focusing is consistent, and continuing to execute step S44 to perform region contraction; otherwise, the focusing is determined to be inconsistent, and multi-focus tracking is started.
  5. 5. The method for predicting the health of an industrial device based on multi-source data fusion of claim 4, wherein the multi-focus tracking is performed by: sub-scan regions are created for the plurality of high abnormality index units currently identified, respectively, and step S4 is executed in parallel for each sub-scan region.
  6. 6. The method for predicting the health of industrial equipment based on multi-source data fusion according to claim 1 or 5, wherein in the step S1, the electrical performance monitoring data includes high-frequency current signals, voltage harmonics and reactive power curves; the insulation state monitoring data comprise the content of dissolved gas in oil, partial discharge signals and top oil temperature; Wherein the content of dissolved gas in the oil is monitored by an on-line oil chromatograph, and the partial discharge signal is monitored by a partial discharge sensor.
  7. 7. The method for predicting the health of industrial equipment based on multi-source data fusion according to claim 3, wherein in step S41, the local abnormality index of each grid is calculated, and the specific process of identifying the kth hot spot unit is as follows: S441, spatial positioning monitoring data based on the kth scanning is obtained, wherein the spatial positioning monitoring data comprises one or more of ultrasonic signal amplitude, ultrahigh frequency signal energy and infrared temperature measurement data; s442, calculating a comprehensive local abnormality index by weighted summation for each grid cell in the current scanning area, wherein the calculation formula is as follows: ; Wherein, the As a local abnormality index, Is a normalized value of the amplitude of the ultrasonic signal, Is the normalized value of the energy of the ultrahigh frequency signal, Is a normalized value of the temperature value, , , Is a weight coefficient, and + + =1; S443, comparing the local abnormality indexes of all grids, and identifying the grid unit with the highest local abnormality index and exceeding a preset activity threshold as the kth hot spot unit.
  8. 8. The method for predicting the health of industrial equipment based on multi-source data fusion according to claim 1, wherein the step of obtaining the short-term variation trend value is as follows: S31, acquiring a history sequence of each parameter in the electrical performance monitoring data in a latest time window T, and respectively calculating standard deviations of the history sequence; S32, taking the parameter with the largest standard deviation as a key electrical parameter; S33, performing linear regression analysis on the historical sequence corresponding to the key electrical parameters, and calculating the slope value of the key electrical parameters; S34, taking the absolute value of the slope value as the numerical value of the short-term change trend.
  9. 9. The method for predicting the health of industrial equipment based on multi-source data fusion according to claim 1, wherein the pre-trained health assessment model is a machine learning model based on a gradient lifting decision tree; The health evaluation model is obtained through historical data training, and the historical data comprise electrical performance monitoring data and insulation state monitoring data corresponding to the normal operation state and various known fault states of the transformer; The integrated health index is a scalar between 0 and 1 for quantifying a continuous health status characterizing the transformer from full health to full failure.

Description

Industrial equipment health state prediction method based on multi-source data fusion Technical Field The invention relates to the field of health monitoring, in particular to an industrial equipment health state prediction method based on multi-source data fusion. Background Industrial equipment, such as transformers and energy storage equipment in power systems, and pumps, fans, compressors and the like in process industry, are core assets for guaranteeing social production and energy safety. The accurate prediction of the health state and the early diagnosis of faults are significant in preventing unplanned shutdown, reducing maintenance cost and avoiding disastrous accidents. In recent years, with the popularization of sensor technology and internet of things (IoT), data-based predictive maintenance (PdM) strategies are gradually replacing traditional periodic maintenance and post-maintenance, and are becoming the main stream direction of industrial equipment health management. Existing predictive maintenance methods can be broadly divided into the following categories: 1. an analysis method based on a single data source; Such methods rely on a single type of sensor data. For example, vibration analysis is often used for fault diagnosis of rotating machinery, oil chromatography (DGA) is a "gold criterion" for judging insulation faults inside transformers, and current signature analysis (MCSA) is used for diagnosing electrical faults such as broken bars, eccentricity and the like of motor rotors. However, such methods have significant limitations in that, first, the failure modes of the industrial equipment are complex and diverse, and a single type of data can only reflect the state of a certain side of the equipment, and lack of a global view angle is extremely prone to misjudgment or missed judgment. For example, vibration data alone may not be effective in distinguishing between mechanical imbalance and electrical faults. Secondly, these methods generally rely on expert experience to set thresholds, and have limited degree of intelligence, and it is difficult to achieve automatic identification of early and weak fault signs. 2. A simple multi-source data parallel monitoring method; In order to obtain more comprehensive information, some systems begin to deploy various sensors and collect various signals of vibration, temperature, current, acoustics and the like in parallel. However, most of the current systems only stay at the level of 'parallel display' or 'independent alarm' of data, and lack a deep information fusion mechanism. The operators need to manually compare alarm information from different systems to comprehensively judge the fault, so that the efficiency is low, personal experience is seriously relied on, and the consistency is poor. 3. Preliminary data fusion attempts; There have been some attempts in the art to fuse multi-source data. For example, feature vectors of various sensors are simply spliced and then input into a classifier to perform state recognition. However, such methods remain static and stiff. The detection strategy (such as sampling frequency and scanning range of the sensor) is usually preset, and cannot be dynamically adjusted according to the real-time state of the device. When the system detects abnormal trend, the system cannot automatically focus resources to conduct deep exploration, and when the equipment state is stable, high-density scanning is continuously conducted, so that great waste of computing resources and communication bandwidth is caused, and the economy of large-scale deployment of the technology on the edge side of limited resources is limited. In view of the above, the prior art lacks a solution capable of adaptively, efficiently and precisely implementing the prediction of the health status of industrial equipment. Disclosure of Invention The invention aims to provide an industrial equipment health state prediction method based on multi-source data fusion, which solves at least one technical problem. The aim of the invention can be achieved by the following technical scheme: the industrial equipment health state prediction method based on multi-source data fusion comprises the following steps: S1, synchronously acquiring electrical performance monitoring data and insulation state monitoring data of a transformer; s2, fusing the characteristics of the electrical performance monitoring data and the insulation state monitoring data, inputting the characteristics into a pre-trained health evaluation model, and calculating to obtain a comprehensive health index; s3, continuously calculating short-term change trend of the electrical performance monitoring data when the industrial equipment does not enter an abnormal state, and dynamically adjusting acquisition frequency or scanning intensity of the insulation state monitoring data according to the numerical value of the short-term change trend; Wherein, the numerical value of the short-term variation t