Machine Learning Network Fault Detection in Wiring Harnesses
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods are inefficient in testing complex electrical networks in wiring harnesses, particularly at high frequencies, due to the superimposition of reflections which obscures individual imperfections in measurement curves, making it difficult to identify errors in complex networks of UTP lines.
Innovation Solution
A method and device utilizing machine learning algorithms to classify network sections as error-free or error-prone, involving training measurement values from reference networks, preprocessing to eliminate data errors, and using two classification systems to identify errors and localize faults in complex electrical networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional measurement methods are used to test complex electrical networks, then the testing process is simple, but the measurement precision deteriorates due to superimposed reflections obscuring individual imperfections
Solution Approach 1:
The patent divides the complex network testing problem into two separate classification tasks: a first classification system that determines whether a network is error-free or error-prone, and a second classification system that identifies the specific error-prone network section. This segmentation of the classification function enables precise fault detection in complex networks by breaking down the overwhelming measurement data into manageable classification stages, thereby improving measurement precision without requiring a single overly complex testing device.
2Measurement precision
If machine learning classification systems are implemented to improve fault detection accuracy, then the measurement precision improves, but the device complexity increases due to training requirements and computational resources
Solution Approach 1:
The patent implements preliminary action by training both classification systems using training measurement values obtained from reference networks before actual testing. The training phase prepares the classification systems in advance, allowing them to automatically classify test measurement values during actual operation without requiring complex real-time computational resources. This preliminary training approach enables high fault localization accuracy while managing device complexity by shifting computational burden to an offline training stage.
3Productivity
If two classification systems are used to localize faults precisely, then the productivity of fault identification improves, but the device complexity increases due to multiple processing stages
Solution Approach 1:
The patent segments the fault identification process into two sequential classification stages: first determining whether a network contains errors, then identifying the specific error-prone section. This segmentation improves productivity by enabling rapid screening of networks through the first classification system, followed by detailed fault localization only for networks that require it. The two-stage approach efficiently handles large numbers of networks while managing system complexity through modular processing stages.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
The present invention discloses a method for testing a network (151, 360, 461) having a number of network sections (A - I, J - Q), in particular in a cable harness having a number of such networks (151, 360, 461), having the following steps of: recording (S1) training measured values (102) for a number of reference networks (150), wherein the reference networks (150) correspond to the network (151, 360, 461) to be tested, preprocessing (S2) the recorded training measured values (102) in order to eliminate data errors in the training measured values (102), training (S3) a first classification system (104, 204) using the training measured values (102), wherein the first classification system (104, 204) is based on at least one algorithm from the field of machine learning and is designed to classify a network (151, 360, 461) either as fault-free or faulty, training (S4) a second classification system (105, 205) using the training measured values (102), wherein the second classification system (105, 205) is based on at least one algorithm from the field of machine learning and is designed to classify a faulty network section (A - I, J - Q) of a network (151, 360, 461), recording (S5) test measured values (107) for the network (151, 360, 461) to be tested, preprocessing (S6) the recorded test measured values (107) in order to eliminate data errors in the training measured values (102), classifying (S7) the network (151, 360, 461) to be tested as fault-free or faulty on the basis of the recorded test measured values (107) using the trained first classification system (104, 204), and classifying (S8) the faulty network section (A - I, J - Q) of the network (151, 360, 461) to be tested using the trained second classification system (105, 205) if the network (151, 360, 461) to be tested was classified as faulty by the trained first classification system (104, 204). The present invention also discloses a corresponding testing device.