CN-122020221-A - Equipment fault prediction and health management method and system based on multi-source data fusion

CN122020221ACN 122020221 ACN122020221 ACN 122020221ACN-122020221-A

Abstract

The invention discloses a device fault prediction and health management method and system based on multi-source data fusion, and relates to the field of system predictive maintenance based on industrial Internet, comprising S1, in the healthy operation stage of a device, synchronously collecting parameters related to the operation of the device in a multi-dimension manner as historical data, and forming a training data set through the historical data; S2, constructing a reference model through a multi-manifold soft comparison learning algorithm based on the training data set, S3, synchronously collecting data streams which run in real time in a manner of keeping the dimension and frequency consistent with the historical data in the real-time running process of the equipment to obtain a real-time sample sequence x t , S4, inputting x t into the reference model to obtain a comprehensive health state dissimilarity value S t , S5, comparing S t with a preset multi-stage early warning threshold value to trigger a corresponding fault processing mode according to a comparison result. The invention can avoid misjudging normal working condition switching as a fault, and remarkably improves the reliability of the system.

Inventors

YIN LINGYU
LI MENGYANG
LI KAICEN
CHEN JING
CHEN HAIPING
LV HAIBING
FAN NAIJI

Assignees

中国工程物理研究院激光聚变研究中心

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (8)

1. The equipment fault prediction and health management method based on multi-source data fusion is characterized by comprising the following steps of: S1, in a healthy operation stage of equipment, synchronously acquiring parameters related to the operation of the equipment in a multi-dimensional manner as a historical healthy operation data set D train , and forming a training data set through the historical healthy operation data set D train ; S2, constructing a reference model capable of describing multiple health conditions of the equipment from D train by a multi-manifold soft contrast learning algorithm based on the training data set; S3, in the real-time operation process of the equipment, maintaining the dimension and frequency consistent with the historical data to synchronously acquire the data flow operated in real time so as to obtain a real-time sample sequence x t ; s4, inputting x t into a reference model to obtain a comprehensive health state dissimilarity value S t ; S5, comparing the health state dissimilarity value S t with a preset multilevel early warning threshold value to trigger a corresponding fault processing mode according to a comparison result; the construction mode of the reference model comprises the following steps: S20, preliminarily dividing a training data set into K sub-manifolds by a Gaussian mixture model unsupervised clustering method, wherein each sub-manifold corresponds to a stable working condition; S21, constructing sample pairs based on each manifold, and designing a depth encoder for mapping the sample pairs to a low-dimensional discriminant space And optimizing each manifold data by adopting a soft contrast learning algorithm, wherein, Representing the depth encoder function itself, the subscript theta represents the set of ownership weights and bias parameters in the corresponding network, The input space is represented by a representation of the input space, Representing the output space.
2. The method for predicting equipment failure and managing health based on multi-source data fusion as set forth in claim 1, wherein in S20, the data distribution of the kth manifold is expressed as: in the above formula, x is a d-dimensional observation data vector, and , Is the model parameter for the kth sub-manifold, J is the number of gaussian mixture components within the current manifold, Is the mixing weight of the jth gaussian component, Representing an average value as Covariance matrix is P () is a probability density function, Parameters representing a given kth healthy sub-manifold When the probability density of the multidimensional data point x is observed.
3. The method for predicting equipment failure and managing health based on multi-source data fusion of claim 1, wherein the multiple gaussian distributions Characterization was performed by the following formula: in the above-mentioned method, the step of, Covariance matrix representing jth gaussian component in kth healthy sub-manifold Is used for the inverse matrix of (a), Representing data point x to the corresponding gaussian component center Is the square of the mahalanobis distance.
4. The method for predicting equipment failure and managing health based on multi-source data fusion of claim 1, wherein in S21, the soft-contrast learning algorithm loses function Characterization was performed by the following formula: In the above-mentioned method, the step of, Is a set of positive sample pairs consisting of data enhanced samples within the same sub-manifold, Is a negative sample set of anchor points x i , s () is cosine similarity, Is used for the temperature super-parameter, For positive pairs of samples from the same healthy sub-manifold, Are pairs of soft negative samples from different but adjacent healthy sub-manifolds.
5. The method for predicting and managing equipment failure based on multi-source data fusion of claim 1, wherein in S1, the history data includes vibration signal, temperature, operation current, noise signal, and servo driving parameters outputted from the equipment controller.
6. The method for predicting equipment failure and managing health based on multi-source data fusion according to claim 1, wherein in S4, the method for obtaining the health state dissimilarity value S t is as follows: S40, the encoder processes the real-time sample sequence x t by the following formula to obtain a corresponding characterization vector Z t : S41, calculating weighted minimum distances from Z t to all healthy manifold centers by the following formula, and taking the weighted minimum distances as health state dissimilarity values: Wherein, the For Z t to kth manifold center Is a mahalanobis distance or an euclidean distance, Is a weight based on healthy confidence of the kth manifold history data.
7. The method for predicting equipment failure and managing health based on multi-source data fusion as set forth in claim 1, wherein in S5, the multi-level early warning threshold includes an attention threshold, a warning threshold, and a danger threshold; In S5, the processing mode is: When the health state dissimilarity value S t continuously exceeds the attention threshold and shows an ascending trend, triggering early warning, and creating an early warning diagnosis list in a database; triggering a maintenance work order when the health state dissimilarity value S t approaches the warning threshold, and recording and generating an emergency maintenance work order containing a fault mode and affected parts; wherein the affected part is positioned by: Wherein, the For the j-th dimension sense signal of the corresponding device component, For the purpose of deviation guiding.
8. The equipment fault prediction and health management system based on multi-source data fusion as claimed in any one of claims 1-7 is characterized by comprising a multi-source data acquisition layer, a health model construction layer, a real-time state evaluation layer and an intelligent early warning and decision layer; The multi-source data acquisition layer comprises a data acquisition module arranged on equipment and a PLC (programmable logic controller) in communication connection with the data acquisition module, wherein the data acquisition module comprises a vibration sensor, a temperature sensor, a current transformer and a noise microphone; The health model construction layer, the real-time state evaluation layer and the intelligent early warning and decision layer are all arranged on the server, and the multi-source data acquisition layer is in communication connection with the server through an industrial Ethernet.

Description

Equipment fault prediction and health management method and system based on multi-source data fusion Technical Field The invention relates to the field of predictive maintenance of systems based on the industrial Internet. More particularly, the invention relates to a device fault prediction and health management method and system based on multi-source data fusion for key devices (such as stackers, AGVs and conveyor lines) in an automated warehouse logistics system. Background The automatic warehouse logistics system is a core link of modern intelligent manufacturing, and continuous and stable operation of core equipment (such as a stacker, an AGV and a conveying line) is important. At present, the maintenance of core equipment in an automatic warehouse logistics system mainly adopts the following two modes: firstly, maintenance is carried out afterwards, namely, the equipment is stopped for maintenance after the equipment is in fault, and the problem is that the production is interrupted, so that huge economic loss is caused. And secondly, regular preventive maintenance, namely maintenance based on fixed time or operation period, regardless of the actual state of the equipment. The mode has two defects, namely insufficient maintenance and possible failure of equipment in a period of short maintenance, excessive maintenance, and the equipment is still detached and replaced when the equipment is in a good state, so that manpower and spare part resources are wasted. Of course, monitoring based on sensor threshold values, such as monitoring motor temperature and alarming when a certain fixed threshold value is exceeded, is also introduced in the prior art, however, the method is still a hysteresis and primary diagnosis rather than prediction, namely, the performance degradation trend of equipment cannot be identified, only alarming when faults are about to occur or occur, and the response time for operation and maintenance personnel is extremely short. In addition, the sensor threshold monitoring adopted in the prior art is mostly single-parameter and single-point monitoring, and the comprehensive analysis and fusion judgment on the multi-dimensional operation state (such as vibration, temperature, current, noise and control parameters) of the equipment are lacked. For example, early bearing wear of the motor may only manifest as a small change in the vibration spectrum, while the temperature has not risen yet, and such potential faults cannot be found by temperature monitoring alone. Therefore, the existing maintenance mode is lag, passive and inaccurate in single parameter monitoring, mainly comprises the following steps of being incapable of early warning of early hidden faults, being incapable of quantifying the decline trend of the health state of equipment, being incapable of providing decision basis for accurate maintenance, and being incapable of realizing early, accurate and prospective fault prediction of warehouse logistics equipment due to unplanned shutdown and resource waste caused by improper maintenance, namely being difficult to support strategic transformation from 'preventive maintenance' to 'predictive maintenance'. Disclosure of Invention It is an object of the present invention to address at least the above problems and/or disadvantages and to provide at least the advantages described below. To achieve these objects and other advantages and in accordance with the purpose of the invention, as embodied and broadly described herein, there is provided an apparatus fault prediction and health management method based on multi-source data fusion, comprising: S1, in a healthy operation stage of equipment, synchronously acquiring parameters related to the operation of the equipment in a multi-dimensional manner as a historical healthy operation data set D train, and forming a training data set through the historical healthy operation data set D train; S2, constructing a reference model capable of describing multiple health conditions of the equipment from D train by a multi-manifold soft contrast learning algorithm based on the training data set; S3, in the real-time operation process of the equipment, maintaining the dimension and frequency consistent with the historical data to synchronously acquire the data flow operated in real time so as to obtain a real-time sample sequence x t; s4, inputting x t into a reference model to obtain a comprehensive health state dissimilarity value S t; S5, comparing the health state dissimilarity value S t with a preset multilevel early warning threshold value to trigger a corresponding fault processing mode according to a comparison result; the construction mode of the reference model comprises the following steps: S20, preliminarily dividing a training data set into K sub-manifolds by a Gaussian mixture model unsupervised clustering method, wherein each sub-manifold corresponds to a stable working condition; S21, constructing sample pairs based on each manifol