Sliding Surface Diagnosis Using Multi-Sensor Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sliding-surface diagnosis technologies rely on individual physical quantities and operator experience, failing to accurately diagnose complex changes in sliding surface conditions and exhibiting significant individual operator variability.
Innovation Solution
A sliding-surface diagnosis apparatus using a machine learning model generated by a machine learning apparatus to diagnose the condition of sliding surfaces based on multiple physical quantities, employing a neural network for supervised learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If diagnosis is based on individual physical quantities (pressure, temperature, vibration), then the diagnosis system is simple, but it cannot accurately diagnose complex changes in sliding surface conditions
Solution Approach 1:
The patent combines multiple physical quantities (pressure, temperature, vibration, etc.) into a unified diagnosis system using machine learning. The machine learning model integrates these diverse parameters to comprehensively assess sliding surface conditions, enabling accurate diagnosis of complex changes that cannot be detected by individual parameters alone.
Solution Approach 2:
The patent introduces a machine learning model as an intermediary between raw physical quantity measurements and diagnosis results. This intermediary processes and synthesizes multiple physical quantities, transforming complex multi-parameter data into accurate diagnostic information about sliding surface conditions.
2Measurement precision
If diagnosis relies on operator experience, then the system is simple to implement, but there are large individual differences between operators
Solution Approach 1:
The patent replaces the human operator's experience-based judgment with a machine learning model. The model learns optimal diagnosis criteria from training data and automatically applies this knowledge to new cases, eliminating individual differences between operators and providing consistent, objective diagnosis results.
Solution Approach 2:
The machine learning model performs self-learning from training data consisting of physical quantity measurements and corresponding diagnosis results. Through this self-service learning process, the model automatically acquires diagnostic capabilities without requiring manual programming of expert knowledge, achieving both consistency and automation.
3Reliability
If multiple physical quantities are used for diagnosis, then comprehensive assessment is achieved, but the diagnosis process becomes complex and time-consuming
Solution Approach 1:
The patent performs preliminary action by training the machine learning model in advance using comprehensive training data that includes multiple physical quantities and corresponding diagnosis results. This pre-training enables the model to rapidly process new measurements without requiring complex real-time analysis, reducing diagnosis time while maintaining reliability.
Solution Approach 2:
The patent transforms multiple physical quantity parameters into a unified diagnostic assessment through the machine learning model. The model learns optimal parameter transformations and relationships during training, enabling efficient processing of multi-parameter data and rapid generation of reliable diagnosis results.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A machine learning apparatus (4) generates a learning model (6) to be used in a sliding-surface diagnosis apparatus (1A) for diagnosing a condition of sliding surfaces of a fixed-side sliding member and a rotation-side sliding member. The machine learning apparatus (4) includes: a learning-data memory (41) configured to store learning data including input data containing at least data on motor current value in a predetermined period, data on contact electric resistance in the predetermined period, and data on vibration (AE wave or acceleration) in the predetermined period; a machine learning section (42) configured to input the learning data to the learning model (6) to cause the learning model (6) to learn a correlation between the input data and diagnostic information of the sliding surfaces; and a learned-model memory (43) configured to store the learning model (6) that has learned by the machine learning section (42).