my account

Process control


This task will create a module, which will integrate measurements with process models for the root cause analysis of 6-sigma process faults in RLW involving 3 steps: The main challenges of the development of the root cause analysis module for RLW operation are: (i) the high dimensionality of the faults space affecting the process; (ii) data heterogeneity (data coming from different sensors -- multisensory data fusion problem); and (iii) the use of different sampling rates during production (combination of in-line and off-line measurements e.g. non-destructive and destructive tests) resulting in some process variables measured for every weld, some for every parts and others for few parts in a batch of parts. The root cause analysis will be conducted in 3 steps:

•fault identification (identifying fault or no fault scenario and magnitude of the fault)

•fault localization (localizing where the fault occurred in the process, i.e., process component); and

•fault isolation (isolating the root cause(s) of the fault).

Fault identification involves identification of the key measurements explaining the weld quality and dimensional & geometrical variation during the joining. The fault isolation method will be conducted by mapping variation pattern(s) of a single error on the joined parts with the process parameter space to obtain fault covering set which minimize a conflict space (a set of process parameters which support/confirm the fault root cause(s) that lead to product dimensional variation).

The fault localization and isolation modules will be based on statistical testing to select fault/no-fault scenarios. The fault localization and fault isolation are based on the proposed Enhanced Piecewise Least Squares (EPLS) approach which searches for a set of components called latent vectors with the search constrained by the response function of the RLW process obtained from the Error models. The proposed EPLS method performs the decomposition of data response-based final product measurements. The EPLS method will have capability to construct multiple faults predictive models for ill-defined RLW systems with large numbers of highly collinear factors. This is very critical for RLW systems with a large number of collinear process parameters. Since the measurement data have very high dimensionality (e.g. surface deformation measured with an optical CMM and ultrasound non-destructive testing); a new approach will be proposed to reduce model dimensionality. The dimensionality reduction will utilize the SMA technique introduced in Task 3.2 to allow modelling not only error of a single product but 6-sigma fault patterns of a batch of products. In essence the developed methodology will allows decomposition of the fault pattern into orthogonal "variation modes".


Web Part Quick Links

Latest News

RLW Navigator is a three year, €3.9M, project funded by the European Commission under the ICT-Factories of the Future programme.  The project has fourteen partners and began in January 2012.

The goal of the project is to develop an engineering platform for an emerging joining technology from the automotive industry, Remote Laser Welding (RLW), that will enable the exploitation of this technology and ultimately support other joining processes.

©2013 RLW Navigator. All rights reserved.