CN-121980326-A - Ship intelligent fault diagnosis method and system based on open label space recognition

CN121980326ACN 121980326 ACN121980326 ACN 121980326ACN-121980326-A

Abstract

The invention discloses a ship intelligent fault diagnosis method and system based on open label space recognition, comprising the following steps of obtaining a multi-source sensor time sequence signal of a ship system and constructing a training sample set, carrying out feature extraction on the training sample set by utilizing a deep neural network model, adopting a central loss function to strengthen the clustering characteristic of the features in the training process, training a classifier on the basis of feature extraction and introducing an open set loss function to form a comprehensive objective function, embedding the features extracted from a sample to be diagnosed and calculating the distance between the features extracted from the sample to be diagnosed and the center of the features of the known fault class, judging unknown faults through comparison of the minimum distance and a preset threshold value, outputting the known fault class or triggering unknown fault alarm according to the judging result, and dynamically expanding and updating a model knowledge base based on the accumulated unknown fault samples. The invention can break through the limitation of the traditional closed set assumption, effectively identify the unknown fault type and realize the self-adaptive learning and continuous optimization of the ship fault diagnosis system.

Inventors

YUAN DONGLEI
LV WEIQIAN
LI XIAOCHI
MA FENG
SUN JIE
CHEN CHEN
LU BILIANG

Assignees

河南港航集团有限公司

Dates

Publication Date: 20260505
Application Date: 20251202

Claims (10)

1. The intelligent fault diagnosis method for the ship based on the open label space identification is characterized by comprising the following steps of: Acquiring a multi-source sensor time sequence signal of a ship system and constructing a training sample set; extracting features of the training sample set by using a deep neural network model to obtain distinguishing embedded features of the sample; Training a classifier and introducing an open set loss function as a part of a comprehensive objective function of model training on the basis of the feature extraction, wherein the comprehensive objective function is formed by the open set loss function and the central loss function together; Extracting the characteristic embedding of the sample to be diagnosed, and calculating the distance between the sample to be diagnosed and the characteristic centers of all known fault categories; outputting a corresponding fault class if the known fault is judged, triggering an alarm if the unknown fault is judged, and dynamically expanding and updating a fault knowledge base of the model based on accumulation of the unknown fault samples.
2. The intelligent fault diagnosis method for a ship based on open label space recognition according to claim 1, wherein the acquiring the multi-source sensor timing signal of the ship system and constructing the training sample comprises: acquiring multisource sensor signals of at least one system of a ship electric propulsion system, cabin power equipment, an energy conversion device and an auxiliary electromechanical system; segmenting the signal into sample segments of length L by a sliding time window mechanism; And carrying out normalization processing on the sample fragments to obtain a training sample set of the known fault labels.
3. The intelligent fault diagnosis method for the ship based on the open label space identification according to claim 2, wherein the multi-source sensor signals comprise current, voltage, rotating speed, vibration and temperature.
4. The intelligent fault diagnosis method for the ship based on the open label space recognition according to claim 1, wherein the deep neural network model sequentially comprises: The one-dimensional convolution layer and multi-layer residual convolution module is used for extracting local time sequence features; the channel attention mechanism is used for carrying out channel weight recalibration on the convolution characteristics; the self-attention mechanism is used for modeling global dependency of the time dimension of the feature; and the multi-scale pooling and cascading layer is used for fusing the multi-scale characteristics and outputting the distinguishing embedded characteristics.
5. The intelligent fault diagnosis method for the ship based on the open label space recognition according to claim 1, wherein the characteristic clustering loss is a central loss function, and the calculation mode is as follows: Wherein, the For the feature embedding of the sample, Represent the first The feature center of the class.
6. The intelligent fault diagnosis method for the ship based on the open label space recognition according to claim 1, wherein the open set regularization term is an open set loss, and the calculation mode is as follows: Wherein, the Is cross entropy loss; For the open set penalty term, when the sample confidence is below the threshold Punishment is carried out at the time; Is a weight factor.
7. The intelligent fault diagnosis method for the ship based on the open label space recognition according to claim 1, wherein the distance between the intelligent fault diagnosis method and the feature center of all known fault categories is calculated by adopting a distance measurement mode of Euclidean distance, and a calculation formula is as follows:
8. the intelligent fault diagnosis method for the ship based on the open label space recognition according to claim 1, wherein the clustering analysis of the samples of the unknown faults specifically adopts a DBSCAN density clustering algorithm.
9. The intelligent fault diagnosis method for the ship based on the open label space recognition according to claim 1, wherein the fault knowledge base of the dynamic expansion and update model is realized by re-optimizing model parameters through an incremental learning mechanism.
10. An intelligent fault diagnosis system for a ship based on open label space recognition, characterized in that the intelligent fault diagnosis method for a ship based on open label space recognition according to any one of claims 1 to 9 comprises: The collecting module is used for obtaining the time sequence signals of the multisource sensor of the ship system and constructing a training sample set; The training sample set is subjected to feature extraction by the learning module to obtain the distinguishing embedded features of the samples, and a center loss function is adopted to strengthen the clustering characteristics of similar features in an embedded space in the training process; the training module is used for training the classifier and introducing an open set loss function as a part of a comprehensive objective function for model training on the basis of the feature extraction, wherein the comprehensive objective function is formed by the open set loss function and the central loss function; The diagnosis module extracts the characteristic embedding of the sample to be diagnosed and calculates the distance between the sample to be diagnosed and the characteristic centers of all known faults, and if the minimum distance is larger than a preset threshold value, the sample to be diagnosed is judged to be an unknown fault; The system comprises an extension module, a fault knowledge base, a fault analysis module and a fault analysis module, wherein the extension module outputs a corresponding fault class if judging that a fault is known, triggers an alarm if judging that the fault is unknown, and dynamically extends and updates the fault knowledge base of the model based on accumulation of unknown fault samples.

Description

Ship intelligent fault diagnosis method and system based on open label space recognition Technical Field The invention belongs to the technical field of intelligent monitoring and control of ships, and particularly relates to an intelligent ship fault diagnosis method and system based on open label space identification. Background The green intelligent ship has become an important direction of the development of future ships, and the safe and stable operation of a core system (such as an electric propulsion system) of the intelligent ship is directly related to the sailing reliability of the ship. Currently, ship fault diagnosis techniques mainly rely on intelligent diagnostic models based on deep learning or pattern recognition. However, these methods are generally based on the assumption of a closed tag space, i.e. during the model training phase, all possible fault categories need to be preset. In actual operation of the ship, the electromechanical equipment is in a complex and changeable working environment, and the operation state is easily influenced by various dynamic factors such as load fluctuation, electromagnetic interference, mechanical abrasion, temperature drift and the like. Thus, during the full life of the device, new unknown fault types that were not contained in the training samples are highly likely to occur. Conventional closed hypothesis-based diagnostic models, in the face of these unknown faults, may force classification into a certain known fault type, thereby producing erroneous classification results. More seriously, the model sometimes gives high confidence to the misclassification, so that the system cannot trigger effective alarm or misdiagnosis, which not only remarkably reduces maintenance efficiency, but also brings potential risks to ship navigation safety. Disclosure of Invention In view of the above-mentioned shortcomings in the technical field of intelligent monitoring and control of ships at present, the invention provides an intelligent fault diagnosis method for ships based on open label space identification, which can break through the limit of closed set assumption and realize the intelligent diagnosis method for accurately identifying known faults and adaptively detecting unknown faults under the open label space so as to improve the effects of environmental adaptability and safety guarantee capability of a ship fault diagnosis system. In order to achieve the above purpose, the embodiment of the present invention adopts the following technical scheme: An intelligent fault diagnosis method for a ship based on open label space identification comprises the following steps: Acquiring a multi-source sensor time sequence signal of a ship system and constructing a training sample set; extracting features of the training sample set by using a deep neural network model to obtain distinguishing embedded features of the sample; Training a classifier and introducing an open set loss function as a part of a comprehensive objective function of model training on the basis of the feature extraction, wherein the comprehensive objective function is formed by the open set loss function and the central loss function together; Extracting the characteristic embedding of the sample to be diagnosed, and calculating the distance between the sample to be diagnosed and the characteristic centers of all known fault categories; outputting a corresponding fault class if the known fault is judged, triggering an alarm if the unknown fault is judged, and dynamically expanding and updating a fault knowledge base of the model based on accumulation of the unknown fault samples. According to one aspect of the invention, acquiring the multi-source sensor time sequence signals of the ship system and constructing training samples comprises the steps of acquiring the multi-source sensor signals of at least one system of a ship electric propulsion system, a cabin power device, an energy conversion device and an auxiliary electromechanical system, segmenting the signals into sample fragments with the length of L through a sliding time window mechanism, and carrying out normalization processing on the sample fragments to obtain a training sample set of known fault labels. According to one aspect of the invention, the multi-source sensor signal includes current, voltage, rotational speed, vibration, temperature. According to one aspect of the invention, the deep neural network model sequentially comprises a one-dimensional convolution layer and a multi-layer residual convolution module, a channel attention mechanism, a self-attention mechanism and a multi-scale pooling and cascading layer, wherein the one-dimensional convolution layer and the multi-layer residual convolution module are used for extracting local time sequence features, the channel attention mechanism is used for carrying out channel weight recalibration on the convolution features, the self-attention mechanism is used for carrying out global d