CN-121980434-A - Multi-mode diagnosis auxiliary system and method for marine diesel engine based on large model

CN121980434ACN 121980434 ACN121980434 ACN 121980434ACN-121980434-A

Abstract

The invention provides a marine diesel engine multi-mode diagnosis auxiliary system and a method based on a large model, the system comprehensively uses sensor signals, image data, audio data and text data, the fault state of the diesel engine is accurately identified, intelligently positioned and quickly maintained and suggested, so that the fault diagnosis efficiency and reliability of the marine diesel engine are improved, and the operation and maintenance cost is reduced.

Inventors

XU CHICHENG
HUANG TAO
Zhao Siheng
MA SHIFEI
Shi Yida

Assignees

中国船舶集团有限公司第七一一研究所

Dates

Publication Date: 20260505
Application Date: 20251205

Claims (10)

1. A large model based marine diesel engine multi-modal diagnostic assistance system comprising: The multi-mode data acquisition module is used for acquiring multi-source data of the running state of the diesel engine; the multi-mode feature extraction module is used for extracting feature information of multi-source data by using signal processing and an artificial intelligent algorithm; the multi-mode fusion diagnosis module is used for introducing a retrieval enhancement generation mechanism to realize the collaborative diagnosis capability of fusing multi-mode data and a structured fault diagnosis knowledge base; the intelligent decision recommendation module is used for automatically generating detailed maintenance guidance steps and decisions according to the diagnosis results output by the multi-mode fusion diagnosis module and the maintenance strategies recommended by the knowledge base; and the man-machine interaction feedback optimization module is used for providing a real-time feedback mechanism for maintenance site personnel.
2. The system of claim 1, wherein the multi-modal data collection module obtains multi-source data of the operating state of the diesel engine in real time.
3. The system of claim 1, wherein the multi-modal data collection module periodically obtains multi-source data of the operating conditions of the diesel engine.
4. A system according to claim 2 or 3, wherein acquiring multi-source data of the operating state of the diesel engine comprises: the sensor input comprises a vibration sensor, a rotating speed sensor and a temperature sensor, wherein the vibration sensor, the rotating speed sensor and the temperature sensor are used for respectively measuring the mechanical vibration, the rotating speed change and the temperature change data of key parts of the diesel engine in real time; the image input comprises a thermal imaging camera and an industrial endoscope, and is used for shooting high-definition images of the surface thermal anomalies and internal key components of the diesel engine so as to find mechanical problems including cracks, wear, blockage and leakage; The time sequence audio input is that the sound signal of the diesel engine during operation is recorded in real time through a high-sensitivity microphone or an audio acquisition system so as to detect abnormal sound phenomena including bearing abrasion, knocking and abnormal cylinder sound; the text input comprises daily maintenance inspection records, equipment operation logs and maintenance work order records, and the historical operation and maintenance conditions of the equipment are deeply understood through text information.
5. The system of claim 1, wherein extracting feature information of the multi-source data using signal processing and artificial intelligence algorithms comprises: Extracting time domain, frequency domain and time-frequency domain characteristics by adopting a correlation algorithm to identify mechanical vibration abnormality, bearing abrasion and dynamic performance change, wherein the correlation algorithm comprises a fast Fourier transform method, a wavelet analysis method and an envelope analysis method; the image data comprises the steps of automatically identifying equipment cracks, leakage, blockage or thermal abnormality areas by utilizing a convolutional neural network, target detection and image segmentation algorithm, and rapidly positioning external mechanical damage points in a diesel engine; extracting sound frequency spectrum characteristics by adopting a Mel frequency cepstrum coefficient and a frequency spectrum analysis technology, and identifying abnormal noise modes occurring in the running process of the diesel engine; And the text data is subjected to deep semantic analysis by adopting a natural language processing technology, key semantic information related to faults in records is accurately extracted, the natural language comprises BERT and a large language model, and the key semantic information related to the faults comprises frequently-occurring fault parts, historical maintenance records and technician feedback information.
6. The system of claim 1, wherein introducing a search enhancement generation mechanism to achieve collaborative diagnostic capability that fuses multimodal data with a structured fault diagnosis knowledge base comprises: Firstly, integrating feature vectors extracted by all modes to form a unified device running state feature expression; then, based on the current multi-mode feature expression of the equipment, retrieving historical cases, fault modes or standard maintenance guide information similar to the current features from a fault diagnosis knowledge base; further, the structural knowledge retrieved by the knowledge base and the characteristics fused in real time are input into a deep fusion network together to perform joint analysis and fault reasoning, and the specific fault types, severity and possible influences possibly existing in the current diesel engine are intelligently judged.
7. The system of claim 6, wherein the structured knowledge base comprising failure modes, failure manifestations, root cause analysis, maintenance procedures, historical maintenance cases is constructed, and is established by using domain expert knowledge and historical maintenance records, and updated in real time by a man-machine interaction feedback module, so that the diagnosis accuracy is continuously improved.
8. The system of claim 1, wherein automatically generating detailed repair guidance steps and decisions based on the diagnosis results output by the multi-modality fusion diagnosis module in combination with the knowledge base recommended repair strategies comprises: The system automatically generates a standardized maintenance flow and operation steps; the recommended scheme is pushed to field technicians through the intelligent terminal to guide the field efficient maintenance operation.
9. The system of claim 1, wherein providing a real-time feedback mechanism for service field personnel comprises: the field technicians can describe the field maintenance condition and special requirements through voice or text; The system automatically receives, understands and analyzes feedback information and dynamically adjusts the diagnosis model and knowledge base content; the on-site feedback information is continuously recorded and accumulated, the diagnosis knowledge base is automatically updated, and the accurate adaptation capability of the system in an actual scene is improved.
10. The marine diesel engine multi-mode diagnosis auxiliary method based on the large model is characterized by comprising the following steps of: 1) In-situ multimodal data acquisition stage When abnormal phenomenon occurs in the running process of the marine diesel engine, the system starts the multi-mode data acquisition module firstly: The vibration sensor acquires vibration signals of key parts of the diesel engine; The current running rotating speed and temperature of the diesel engine are monitored in real time by the rotating speed and temperature sensor; the thermal imaging instrument and the endoscope acquire thermal imaging images of key components of the diesel engine and high-definition photos of internal parts; The audio acquisition equipment records sounds emitted in the running process of the diesel engine in real time; a technician inputs the current inspection condition, log description and maintenance work order text information through a maintenance platform; 2) Multimodal feature extraction stage The vibration signal is analyzed and extracted through FFT and wavelet to obtain frequency spectrum and time domain characteristics, and bearing abrasion and abnormal impact phenomena are identified; The thermal imaging and the endoscope image automatically identify part cracks, leakage points and abnormal hot spot areas through a CNN target detection algorithm; the audio signal is extracted through MFCC characteristics, and abnormal mechanical noise is detected by combining voiceprint recognition; The text information is subjected to semantic analysis through a BERT model, and core information of key maintenance records and fault description is automatically extracted; 3) Multi-modal fusion analysis and diagnosis phases The system fuses the modal characteristics to generate a comprehensive characteristic expression of the fault state of the diesel engine; according to the comprehensive feature expression, automatically retrieving a historical fault case with highest similarity with the current abnormality from a structured fault diagnosis knowledge base; the structural knowledge retrieved from the knowledge base is utilized to assist the current feature analysis, so that the bearing abrasion fault of the current diesel engine is further defined, and the fault positioning precision and speed are obviously improved; 4) Intelligent decision recommendation stage The system combines the historical maintenance cases, expert knowledge patterns and fusion diagnosis results to automatically generate a detailed maintenance proposal flow; the recommended flow is directly pushed to site maintenance personnel through an intelligent decision platform to guide site efficient maintenance operation; 5) Human-computer interaction feedback and optimization stage In the actual operation process, maintenance personnel feed back special conditions in actual maintenance through natural language, the system receives and automatically analyzes feedback information in real time, dynamically adjusts the current maintenance operation scheme, corrects and updates knowledge base contents and diagnosis model parameters, thereby realizing real-time optimization of the knowledge base contents and the diagnosis model, enabling the system to continuously adapt to site conditions, and gradually improving accuracy and adaptability of future fault diagnosis and decision recommendation.

Description

Multi-mode diagnosis auxiliary system and method for marine diesel engine based on large model Technical Field The invention belongs to the technical field of intelligent operation and maintenance of ship power systems, and particularly relates to a multi-mode diagnosis auxiliary system and method for a ship diesel engine based on a large model. Background With the increasing complexity of marine technology, fault diagnosis and maintenance of diesel engines are increasingly dependent on various data sources, such as vibration, temperature, image, sound, and the like. However, most of the prior art rely on a single data source (e.g., sensor data or manual recording), and it is difficult to comprehensively and accurately diagnose faults. The traditional fault diagnosis system mainly depends on manual experience, is low in efficiency and is easily influenced by human factors, and complex mechanical faults cannot be identified in time. Disclosure of Invention In order to overcome the defects of single information, insufficient diagnosis precision, delayed diagnosis response and high dependence on expert experience in the traditional marine diesel engine fault diagnosis means, the invention provides a marine diesel engine multi-mode diagnosis auxiliary system based on a large model, which comprises the following components: The multi-mode data acquisition module is used for acquiring multi-source data of the running state of the diesel engine; the multi-mode feature extraction module is used for extracting feature information of multi-source data by using signal processing and an artificial intelligent algorithm; the multi-mode fusion diagnosis module is used for introducing a retrieval enhancement generation mechanism to realize the collaborative diagnosis capability of fusing multi-mode data and a structured fault diagnosis knowledge base; the intelligent decision recommendation module is used for automatically generating detailed maintenance guidance steps and decisions according to the diagnosis results output by the multi-mode fusion diagnosis module and the maintenance strategies recommended by the knowledge base; and the man-machine interaction feedback optimization module is used for providing a real-time feedback mechanism for maintenance site personnel. Further, the multi-mode data acquisition module acquires multi-source data of the running state of the diesel engine in real time. Further, the multi-mode data acquisition module periodically acquires multi-source data of the running state of the diesel engine. Further, acquiring multi-source data of the operating state of the diesel engine includes: the sensor input comprises a vibration sensor, a rotating speed sensor and a temperature sensor, wherein the vibration sensor, the rotating speed sensor and the temperature sensor are used for respectively measuring the mechanical vibration, the rotating speed change and the temperature change data of key parts of the diesel engine in real time; the image input comprises a thermal imaging camera and an industrial endoscope, and is used for shooting high-definition images of the surface thermal anomalies and internal key components of the diesel engine so as to find mechanical problems including cracks, wear, blockage and leakage; The time sequence audio input is that the sound signal of the diesel engine during operation is recorded in real time through a high-sensitivity microphone or an audio acquisition system so as to detect abnormal sound phenomena including bearing abrasion, knocking and abnormal cylinder sound; the text input comprises daily maintenance inspection records, equipment operation logs and maintenance work order records, and the historical operation and maintenance conditions of the equipment are deeply understood through text information. Further, extracting characteristic information of the multi-source data using signal processing and artificial intelligence algorithms includes: Extracting time domain, frequency domain and time-frequency domain characteristics by adopting a correlation algorithm to identify mechanical vibration abnormality, bearing abrasion and dynamic performance change, wherein the correlation algorithm comprises a fast Fourier transform method, a wavelet analysis method and an envelope analysis method; the image data comprises the steps of automatically identifying equipment cracks, leakage, blockage or thermal abnormality areas by utilizing a convolutional neural network, target detection and image segmentation algorithm, and rapidly positioning external mechanical damage points in a diesel engine; extracting sound frequency spectrum characteristics by adopting a Mel frequency cepstrum coefficient and a frequency spectrum analysis technology, and identifying abnormal noise modes occurring in the running process of the diesel engine; And the text data is subjected to deep semantic analysis by adopting a natural language processing technology, key semantic informatio