CN-121981698-A - Marine diesel engine interactive intelligent agent device based on cooperation of size models and temporary repair method

CN121981698ACN 121981698 ACN121981698 ACN 121981698ACN-121981698-A

Abstract

The embodiment of the invention provides a marine diesel engine interactive intelligent agent device based on size model cooperation and an urgent repair method, aiming at solving the problems of rapid diagnosis, emergency decision and urgent repair guidance when a marine diesel engine suddenly fails. The device comprises a large language model as a core and is responsible for task decomposition, decision generation and information integration, a professional small model performs optimization analysis on tasks such as fault location, emergency decision, spare part replacement and time prediction, an interactive intelligent agent core module supports voice and text input and provides real-time feedback, a multi-mode data real-time fusion module integrates various data sources and dynamically distributes weights, and a knowledge graph and reasoning engine module provides accurate rush-repair guidance. By innovatively carrying out deep coordination on a large language model serving as a core and a plurality of professional small models, rapid diagnosis, emergency decision and temporary repair guidance of sudden faults of the marine diesel engine in the offshore navigation process are realized, and the blank of the intelligent temporary repair technology of the marine diesel engine is filled.

Inventors

MA SHIFEI
LI TINGTING
Zhao Siheng
Shi Yida
JIA SHULI

Assignees

中国船舶集团有限公司第七一一研究所

Dates

Publication Date: 20260505
Application Date: 20251203

Claims (10)

1. The marine diesel engine interactive intelligent agent device based on the cooperation of the size models is characterized in that the composition modules comprise an interactive intelligent agent core, a professional small model cooperation tool chain, a multi-mode data real-time fusion module, an urgent repair knowledge graph and reasoning engine and a man-machine cooperation urgent repair interface, wherein the modules cooperatively realize rapid diagnosis, emergency decision and personalized urgent repair guidance of sudden faults of the marine diesel engine; the interactive intelligent agent core takes a large language model as a base, and has the functions of emergency identification and priority judgment, multi-constraint emergency repair scheme generation, dynamic risk assessment and scheme adjustment through the fine adjustment and knowledge injection in the field of the temporary repair of marine diesel engines; the professional small model cooperative tool chain comprises a fault quick positioning small model, an emergency treatment decision small model, a spare part alternative scheme generation small model and a rush repair time prediction small model, and is scheduled by the interactive intelligent agent core to output a professional subdivision decision result; The multi-mode data real-time fusion module integrates equipment states, environmental conditions, personnel states and resource inventory multi-source heterogeneous data, and dynamically distributes data source importance by adopting an attention mechanism weighting fusion algorithm; the temporary repair knowledge graph and the reasoning engine comprise structured repair knowledge, and path reasoning and knowledge discovery are realized through a graph neural network; The man-machine cooperative rush-repair interface supports voice interaction, remote expert connection and feedback output.
2. The interactive agent device for marine diesel engine based on cooperation of size model according to claim 1, wherein the emergency situation recognition and priority determination of the interactive agent core calculates the fault emergency level by a multi-parameter quantization formula, and the multi-parameters include fault influence range, equipment operation state, navigation environment risk, personnel safety risk.
3. The marine diesel interactive agent device based on size model collaboration according to claim 1, wherein in the professional small model collaboration tool chain: The fault quick positioning small model is based on a lightweight deep learning network, inputs diesel engine sensor data, and outputs fault types and corresponding probability values; the emergency treatment decision small model adopts a reinforcement learning algorithm, and aims at minimizing fault expansion risk and maximizing emergency repair feasibility, and an optimal emergency operation sequence is output; The spare part alternative scheme generating small model provides spare part alternative scheme and corresponding adaptation risk assessment result based on the component compatibility rule base and the similarity algorithm; the rush-repair time prediction small model adopts a time sequence prediction algorithm, and combines fault complexity, personnel skill level, resource in-place condition and environment constraint factors to output rush-repair time consumption and confidence intervals.
4. The interactive intelligent agent device for marine diesel engine based on cooperation of size model according to claim 1, wherein the attention mechanism weighting fusion algorithm of the multi-mode data real-time fusion module is used for supporting subsequent fault diagnosis and decision by dynamically calculating the correlation coefficient between each data source and the current fault scene, distributing differential weight, and outputting fused accurate characteristic representation.
5. The marine diesel engine interactive agent device based on size model cooperation according to claim 1, wherein the critical repair knowledge graph comprises structural knowledge such as fault-symptom-cause-solution association relation, repair standard operation sequence, spare part compatibility rule, resource constraint processing rule, safety operation specification and the like, and supports the realization of knowledge dynamic update through incremental learning.
6. The marine diesel engine interactive agent device based on size model cooperation according to claim 1, wherein the reasoning engine is constructed based on a graph neural network GNN, and fault investigation path reasoning, scheme optimization reasoning under multiple constraint conditions and knowledge discovery of unknown faults are realized by mining hidden relations among entities in a knowledge graph.
7. The interactive intelligent agent device for the marine diesel engine based on the cooperation of the size model according to claim 1, wherein the man-machine cooperation first-aid repair interface is adapted to a marine noisy environment and has a voice noise reduction processing function; The method supports text input and voice interaction dual-mode operation, and the output form comprises fault tree graphical display, three-dimensional equipment model disassembly guidance, rush-repair step text description and a remote expert video link interface.
8. The marine diesel interactive agent device based on size model collaboration according to claim 1, wherein the interactive agent core and professional small model collaboration tool chain adopts a collaborative mechanism of task decomposition-dynamic invocation-result fusion: the interactive intelligent agent core receives the fault request and then disassembles the fault request into subdivision tasks, and dispatches corresponding professional small models according to task types; and collecting output results of all small models, and combining the temporary repair knowledge graph with the reasoning conclusion of the reasoning engine to generate a unified and feasible first-aid repair scheme through semantic understanding and logic integration of the large language model.
9. A marine diesel engine temporary repair method based on size model cooperation, characterized in that the marine diesel engine interactive agent device based on size model cooperation as claimed in any one of claims 1-8 is adopted, comprising the following steps: Step S1, a multi-mode data real-time fusion module collects and integrates equipment state, environmental condition, personnel state and resource inventory multi-source data of a marine diesel engine, and a characteristic representation is output through an attention mechanism weighting fusion algorithm; step S2, the interactive agent core receives the fused characteristic data, invokes an emergency situation recognition function and calculates the emergency level of the fault through a multi-parameter formula; Step S3, according to the emergency level of the fault, the interactive agent core disassembles tasks, dispatches the corresponding small models in the professional small model cooperative tool chain, and outputs a fault positioning result, an emergency operation sequence, a spare part replacement scheme and a rush repair time prediction result respectively; s4, generating a fault investigation path and constraint processing suggestions by inference through a graph neural network based on fault information and a small model output result by the critical repair knowledge graph and inference engine; Step S5, integrating the small model output result and the reasoning conclusion by the interactive intelligent agent core, generating a personalized rush-repair scheme under the multi-constraint condition, and feeding back to a user through a man-machine cooperative rush-repair interface after dynamic risk assessment confirmation; And S6, dynamically adjusting the emergency repair scheme according to the user interaction feedback and the real-time data updating until the fault repair is completed.
10. The method according to claim 9, wherein the dynamic risk assessment in the step S5 is implemented by using a hierarchical analysis method, an assessment index system is constructed from four dimensions of security risk, time risk, cost risk and effect risk, a solution risk value is calculated, and if the risk value exceeds a preset risk threshold value, the step S3 is returned to reschedule the small model optimization solution.

Description

Marine diesel engine interactive intelligent agent device based on cooperation of size models and temporary repair method Technical Field The application relates to the technical field of intelligent operation and maintenance of a ship cabin and artificial intelligence, in particular to a marine diesel engine interactive intelligent body device based on cooperation of a size model and an urgent repair method. Background The intelligent scheduling capability of the cabin power equipment directly influences the feasibility and reliability of the ship navigation plan for the autonomous navigation ship scene. Conventional cabin power plant scheduling schemes generally rely on fixed rotation rules or manual experience, and are prone to causing long-term unbalanced operation of power units, i.e., parts of the units continue to operate under high load, while standby equipment is idle for a long time. On the one hand, the static scheduling strategy can lead to the degradation of the accelerating performance of the main power equipment due to overuse and obviously increase the risk of sudden power interruption, and on the other hand, the redundant resources can not exert the design efficacy through alternate operation, thereby not only weakening the overall robustness of the system, but also being forced to shorten the maintenance period due to the concentrated loss of the local equipment and limiting the sustainability of the sailing plan. The existing cabin power equipment control strategy lacks of adaptability and flexibility, and cannot meet the long-endurance range operation requirement of the autonomous navigation ship. Marine diesel engines, which are used as hearts of ships, often face serious challenges such as incapacitation of harbor maintenance, absence of professional technicians, limited maintenance resources, and the like once they fail during long-period sailing. The traditional marine diesel engine fault treatment mainly relies on the empirical judgment of crews and paper manual review, and has the following key problems in emergency: Temporary rush repair decision hysteresis, namely when a shipman faces a sudden fault, the shipman needs to consult a large number of technical manuals, contact shore-based specialists or wait for guidance of professional technicians, the decision period is long, and the optimal rush repair time is missed. The multi-source information integration is difficult, the temporary rush repair needs to comprehensively consider multi-dimensional information such as fault types, sea condition, spare part inventory, crewman skills, tool availability, navigation plans and the like, and the traditional method is difficult to integrate and analyze rapidly. The acquisition of expertise is limited, in that during offshore navigation, crew is difficult to obtain real-time technical support, especially in the face of complex or rare faults, and there is a lack of effective intelligent decision tools. The personalized defect of the emergency repair scheme is that the existing fault diagnosis system is mostly designed in a universal mode, and a personalized temporary emergency repair scheme cannot be provided according to the equipment configuration, the navigation state and the emergency degree of a specific ship. In recent years, large Language Models (LLMs) have demonstrated strong capabilities in terms of complex reasoning and multitasking coordination, while professional small models have significant advantages in terms of accurate analysis in specific fields. However, no technical solution for applying the deep fusion of the two to the temporary rush repair scene of the marine diesel engine exists at present. The existing intelligent ship operation and maintenance system mainly focuses on conventional monitoring and preventive maintenance and lacks intelligent temporary rush repair decision support in case of emergency faults. Therefore, development of an intelligent body device special for a temporary rush-repair scene of a marine diesel engine is needed, multi-source information can be integrated rapidly under emergency conditions, accurate fault diagnosis and personalized rush-repair guidance are provided, and timely and effective technical support can be ensured when a marine ship encounters a diesel engine fault. Disclosure of Invention The invention aims to provide a marine diesel engine interactive intelligent agent device based on size model cooperation, which solves the problems of delayed temporary rush-repair decision, difficult multi-source information integration, limited acquisition of professional knowledge, insufficient individuation of a rush-repair scheme and the like in the prior art. The device realizes rapid diagnosis, emergency decision and temporary rush repair guidance of sudden faults of the marine diesel engine in the offshore navigation process by innovatively carrying out deep cooperation on a Large Language Model (LLM) serving as a core Agent and a plura