Adaptive Rule Engine for Rolling Stock Maintenance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In remote maintenance of rolling stocks, dispatchers face inefficiencies due to high volumes of diagnostic messages and sensor data, requiring cumbersome manual filtering and analysis, with existing business rules or statistical models often failing to accurately group information, leading to extra work and potential missed or false events.
Innovation Solution
The implementation of an adaptive rule engine that uses machine learning to update and refine existing rules based on dispatcher feedback, applying supervised learning to differentiate between valid and invalid events, thereby automating the adaptation of rule sets for more accurate event identification and maintenance decision-making.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual filtering and analysis of diagnostic messages is used, then dispatchers can identify events requiring maintenance, but the workload increases significantly with large fleets
Solution Approach 1:
The system enables self-service by automatically generating events from diagnostic messages using machine learning models, eliminating the need for manual filtering and analysis by dispatchers. The system serves itself by continuously learning from dispatcher feedback and improving its own event identification accuracy over time.
Solution Approach 2:
The patent replaces the mechanical manual filtering process with an automated machine learning system. Instead of dispatchers manually analyzing diagnostic messages, the system uses trained models to automatically identify events, substituting human mechanical work with automated intelligent processing.
2Extent of automation
If existing business rules or statistical models are used to group diagnostic information, then some automation is achieved, but accuracy is insufficient leading to extra work for dispatchers
Solution Approach 1:
The system implements feedback by using dispatcher actions (acceptance or discarding of events) as training data to continuously improve the machine learning models. This closed-loop feedback mechanism allows the system to learn from errors and progressively enhance event grouping accuracy while maintaining automation.
Solution Approach 2:
The patent applies dynamics by making the rule set adaptive rather than static. The machine learning models are continuously updated based on dispatcher feedback, allowing the system to dynamically adjust its event identification criteria to improve accuracy over time while maintaining high automation levels.
3Measurement precision
If supervised machine learning is applied to learn new rules from dispatcher feedback, then rule accuracy improves, but system complexity increases
Solution Approach 1:
The system manages its own complexity through self-service mechanisms where the machine learning framework automatically trains and updates models using dispatcher feedback. This self-managing approach handles the complexity internally without requiring proportional increases in operational complexity for users.
Solution Approach 2:
The patent introduces machine learning models as intermediaries between raw diagnostic messages and dispatcher decision-making. These models act as intelligent mediators that translate complex data patterns into actionable events, managing system complexity by encapsulating sophisticated processing within the intermediary layer while presenting simplified outputs to users.
Data Source
AI summary
Adaptive remote maintenance of rolling stocks is provided by machine-learning (28) of rules. Existing rules or models are automatically updated. Machine learning (28) is applied to establish a more efficient rule set. Rules may be replaced, generalized, or otherwise adapted based on interaction (26) by the dispatchers with the results of the current rules. The acceptance or discarding of an event by the dispatcher is used as a ground truth for supervised machine learning (28) of a new rule. The machine learning (28) uses user feedback to update the rule set.