CN-122020416-A - Port ship loader dynamic monitoring and multisource information fusion fault diagnosis method

CN122020416ACN 122020416 ACN122020416 ACN 122020416ACN-122020416-A

Abstract

The invention relates to the field of fault diagnosis, in particular to a method for diagnosing a port ship loader by dynamically monitoring and fusing multisource information, which comprises the following steps: and embedding physical constraint into the input high-frequency original data through a depth diagnosis module, generating a theoretical signal, and determining a fault problem in a mode of constructing a loss function. Considering possible causal connections from the past to the present, a causal reasoning module constructs a full connection timing diagram, deletes false correlations through conditional independence verification, obtains a directed causal diagram, and traces back fault propagation paths. The fault components identified by the depth diagnosis module are mapped to the fault propagation paths, and finally determined fault problems are judged based on a consistency verification method. The invention overcomes the excessive dependence of the traditional data-driven fault diagnosis method on priori data by embedding physical constraint in the diagnosis process and combining causal reasoning, and performs the participation of fault diagnosis by combining a physical mechanism.

Inventors

LV QIANG
CUI JIABIN
SONG YUAN
HUO JIDONG
CHANG RONGPENG
LI LAIQIANG
ZHAO JUNRUI
Bian jinxiang

Assignees

国能(天津)港务有限责任公司
中国交通信息科技集团有限公司

Dates

Publication Date: 20260512
Application Date: 20251230

Claims (10)

1. A port ship loader dynamic monitoring and multisource information fusion fault diagnosis method is characterized by comprising the following steps: S10, based on physical loss and data loss in the equipment operation and data transmission process, embedding physical constraint into input high-frequency original data through a depth diagnosis module, generating a theoretical signal, and determining a fault problem in a mode of constructing a loss function; s20, considering possible causal connection from the past to the present, constructing a full-connection time sequence diagram by a causal reasoning module, deleting false association through condition independence test to obtain a directed causal diagram, and tracing a fault propagation path; S30, mapping the fault components identified by the depth diagnosis module to a fault propagation path, and judging the finally determined fault problem based on a consistency check method.
2. The method for diagnosing the fault by combining dynamic monitoring and multi-source information of the port shipment machine according to claim 1, wherein the depth diagnosis module comprises a feature extraction module and a loss calculation module; The feature extraction module comprises a sub-network for extracting depth space-time features from high-frequency original data, the sub-network at least comprises a one-dimensional convolution layer and a bidirectional long-short-time memory network, and the loss calculation module at least comprises a fully-connected sub-network for constructing a loss function and determining a fault problem.
3. The method for diagnosing a fault in dynamic monitoring and multisource information fusion of a port loader according to claim 2, wherein the step S10 comprises the following steps: S11, extracting depth space-time features from high-frequency original data through a feature extraction module; s12, estimating equivalent physical data of the ship loader assembly from the depth space-time characteristics through a fully connected sub-network; S13, calculating theoretical fault frequency according to the geometric parameters and the actual measurement parameters of the ship loader assembly, and acquiring theoretical vibration signals; S14, constructing physical loss based on the theoretical vibration signal and physical constraint, constructing data loss based on equivalent physical data and actual measurement data, and combining the data loss into a total loss function; And S15, minimizing the total loss function through a back propagation algorithm, estimating equivalent physical parameters and identifying a fault component.
4. The method for diagnosing the fault of the dynamic monitoring and the multisource information fusion of the port shipment machine according to claim 1, wherein the acquisition of the high-frequency raw data is dependent on the following steps: s111, acquiring multi-source data through a data acquisition module arranged at each component of the ship loader, and preprocessing the data to obtain a multi-dimensional feature vector; s112, constructing a decision function of an OC-SVM model, and performing anomaly judgment on the multidimensional feature vector through the OC-SVM model; and S113, when the abnormality is judged, performing frequency calculation from the data in the time period in which the abnormality occurs, and separating the data with high frequency occurrence as high-frequency original data.
5. The method for diagnosing a fault by combining dynamic monitoring and multi-source information of a port shipment machine according to claim 1, wherein the constructing of the directed causal graph comprises the following steps: S21, constructing a full-connection time sequence diagram by taking each sensor variable as a node in consideration of possible causal connection from the past; S22, dynamically selecting a given condition set by gradually increasing the size of the condition set, verifying whether all adjacent two variables in the time sequence diagram are independent by adopting a bias correlation coefficient, and deleting edges formed by the two adjacent variables in the time sequence diagram if the two adjacent variables are independent; S23, judging edges in the time sequence diagram by adopting an MCI test model to obtain a directed causal diagram.
6. The method for diagnosing a fault by combining dynamic monitoring and multi-source information of a port loader according to claim 1, wherein the method for tracing the fault propagation path in the step S20 comprises the following steps: And S24, locating nodes with no father node or causal strength inflow larger than outflow in the causal graph, marking the nodes as potential root causes, and tracing the directed paths from the nodes where the potential root causes are located to all abnormal nodes to serve as fault propagation paths.
7. The method for diagnosing a fault by combining dynamic monitoring and multi-source information of a port loader according to claim 1, wherein the step S30 of mapping the fault components identified by the depth diagnosis module onto the fault propagation path comprises the steps of: S31, normalizing key information in a fault part and a causal graph of a depth diagnosis module into a form of a quadruple; S32, constructing a mapping table from the fault component of the depth diagnosis module to the causal graph node based on the physical topology knowledge base.
8. The method for diagnosing a fault by combining dynamic monitoring and multi-source information of a port shipment machine according to claim 7, wherein the construction of the physical topology knowledge base comprises the following steps: S321, establishing a mapping rule from each ship loader component to a sensor signal node connected with the ship loader component according to the physical connection relation and the spatial position of the ship loader component and the sensor; s322, the mapping rule is formed into a physical topology knowledge base for system query.
9. The method for diagnosing a fault by combining dynamic monitoring and multi-source information of a port shipment machine according to any one of claims 1-8, wherein the determining of the finally determined fault based on the consistency check method comprises the steps of: S33, performing space, time and physical consistency check on the fault component corresponding to the fault problem determined by the depth diagnosis module and nodes contained in a causal graph traceable fault propagation path; S34, if the two results are consistent, directly outputting a judging result, if the two results are inconsistent, respectively calculating the confidence coefficient, and when the confidence coefficient difference between the two results exceeds a threshold value, outputting the result with higher confidence coefficient as a fault problem, and taking the other result as a to-be-checked item; when the confidence coefficient of the two is close to the threshold value, judging according to the following mode: For space conflict, if other abnormal nodes exist in the causal graph and the root cause is clear, the output result of the causal reasoning module is taken as the reference; for time conflict, the result of the causal reasoning module is used as the reference; for physical conflicts, the results of the depth diagnostic module are in control.
10. The method for diagnosing a fault by combining dynamic monitoring and multi-source information of a port loader according to claim 9, wherein the confidence calculation of the depth diagnosis module comprises the following formula: Wherein, the 、、 The weight coefficients representing the respective terms are represented, To classify the confidence, refer to the probability that the model outputs the fault class, Representing how well the model output matches the data, Refers to the maximum possible value of the physical loss. The confidence computation of the causal diagnostic module includes the following: Wherein, the 、、 The weight coefficients representing the various indices are used, For the significance of the average signal, the average degree of anomaly of the nodes in the causal graph is represented, For the stability of the causal graph structure, expressed as a consistent case of results over multiple runs, For matching with the priori knowledge, it is indicated whether the causal path coincides with the priori knowledge.

Description

Port ship loader dynamic monitoring and multisource information fusion fault diagnosis method Technical Field The invention relates to the field of fault diagnosis, in particular to a port ship loader dynamic monitoring and multisource information fusion fault diagnosis method. Background With the deep integration of industrial automation and informatization technologies, intelligent equipment has become the core of modern industrial scenes such as ports. The port loader is used as a key device for loading and unloading bulk cargos, and the running process of the port loader involves a large number of complex and continuous transferring operations. If the manual operation and maintenance are completely relied on, the efficiency is low, and high labor cost and safety risk are brought. Therefore, the realization of the automatic operation and the intelligent operation and maintenance of the ship loader is important. In the automatic production process, the timely diagnosis and investigation of equipment faults are key links for guaranteeing continuous and safe operation. Traditional fault handling modes rely on automatic production and manual inspection, and after faults occur, technicians can survey the faults on site and manually remove the faults, so that the mode is slow in response, low in efficiency and dependent on personal experience. With the continuous development of informatization technology, intelligent fault diagnosis technology based on data is beginning to be applied. Such methods typically employ various types of sensors deployed on the device, collect multi-source data such as vibration, temperature, current, etc. during operation, and identify anomalies by comparing the real-time monitored data with historical normal state data. However, this type of pure data driven-based approach has the following problems: The diagnosis result is seriously dependent on the working condition and the fault mode covered by the training data. The physical mechanism of the bottom layer of the equipment is not understood, so that the physical root of the fault is difficult to explain, the generalization capability and the diagnosis reliability of the equipment are obviously reduced when the working condition is severely changed or the type of the fault which is not found is encountered, and the obtained conclusion is likely to be unilateral and the real health state of the equipment cannot be comprehensively and deeply reflected. Disclosure of Invention Aiming at the defects existing in the prior art, the invention provides the following technical scheme: a port ship loader dynamic monitoring and multisource information fusion fault diagnosis method comprises the following steps: And S10, based on physical loss and data loss in the equipment operation and data transmission process, embedding physical constraint into input high-frequency original data through a depth diagnosis module, generating theoretical signals, and determining fault problems through a mode of constructing a loss function. And S20, considering possible causal connections from the past to the present, constructing a full-connection timing diagram by a causal reasoning module, deleting false correlations through a conditional independence test, obtaining a directed causal diagram, and tracing back a fault propagation path. S30, mapping the fault components identified by the depth diagnosis module to a fault propagation path, and judging the finally determined fault problem based on a consistency check method. As an improvement of the technical scheme, the depth diagnosis module comprises a feature extraction module and a loss calculation module. The feature extraction module comprises a sub-network for extracting depth space-time features from high-frequency original data, the sub-network at least comprises a one-dimensional convolution layer and a bidirectional long-short-time memory network, and the loss calculation module at least comprises a fully-connected sub-network for constructing a loss function and determining a fault problem. As an improvement of the above technical solution, the step S10 includes the following steps: And S11, extracting depth space-time features from the high-frequency original data through a feature extraction module. And S12, estimating equivalent physical data of the ship loader assembly from the depth space-time characteristics through the fully connected sub-network. And S13, calculating theoretical fault frequency according to the geometric parameters and the actual measurement parameters of the ship loader assembly, and acquiring theoretical vibration signals. And S14, constructing physical loss based on the theoretical vibration signal and physical constraint, constructing data loss based on equivalent physical data and actual measurement data, and combining the data loss into a total loss function. And S15, minimizing the total loss function through a back propagation algorithm, estimating equivalent phy