Artificial intelligence to detect errors in the production of electronic components

 

How is it possible to design an AI algorithm that can detect and diagnose production line faults through data analytics? Resolving this issue has been the main goal of the “Hirutek” team made up of three researchers at Tekniker, a member of Basque Research and Technology Alliance (BRTA).  

The technology centre’s three experts in data analytics have won the Data Challenge within the context of the leading European predictive maintenance conference, PHME 2021, to be held from June 28 until July 2 in a virtual format.

More specifically, the team submitted an early fault detection system for production lines that tests electronic components.

The solution consists in combining a diagnostic system based on decision trees and a probability propagation algorithm based on Kalman’s filter to diagnose faults and minimise the amount of time required for corrective actions. Thanks to this early fault detection feature, it is possible to minimise risks associated with potentially incipient faults.

The experts behind this solution will present the results of their research in a paper entitled “Divide, Propagate and Conquer: Splitting a Complex Diagnosis Problem for Early Detection of Faults in a Manufacturing Production Line” to be given during the programme of the on-line congress.

In the challenge, moreover, the Hirutek team was awarded a GPU (Graphics Processing Unit) from Nvidia that will allow Tekniker to speed up algorithm counting in future developments.

The excellent results obtained in the challenge by the centre’s analytics group prove the extent to which they are able to position themselves firmly and gain more visibility at an international level in the fields of predictive maintenance and machine data analysis.

Other Tekniker lectures

Tekniker’s participation in PHME 2021 also features a lecture entitled ”A Flexible Data Management System for the Analysis of an ElectroMechanical Actuator on a Test Bench”.

The paper describes an innovative information system developed in the Azure cloud that detects and diagnoses anomalies on the prototype of an electromechanical actuator designed to monitor primary flight surfaces on airplanes.

The system uses the actuator’s operational information compiled by a test bench developed at Tekniker. Anomalies are detected via multivariant statistical process monitoring and an LDA algorithm (Linear Discriminant Analysis) is used for diagnostic purposes.

The PHME conference (Prognostics and Health Management Europe) is the European version of the PHM conference, a benchmark event in the area of predictive maintenance, one of the technological solutions in which Tekniker is a specialist.