knowledge transfer


Quality control for household appliances by on-line evaluation of Mechanical Defects

Submitted by G. Tselentis date 05.07.2001, Service Center


A European manufacturing company of washing machines and dishwashers requires an automatic on-line inspection system to accurately monitor the mechanical characteristic of all their products. Present on-line inspection methods are scarcely effective because they are human operator dependent and not sufficiently reliable. Thus, the main objective is twofold:
a) to have diagnostics of mechanical defects by means of vibrations analysis, thus allowing the percentage of faulty products to decrease
b) to assure and improve the products quality as regards to the emitted noise/ vibrations level. This will happen because the knowledge of defects permits a continuous improvement of the production process.
The system is meant to provide a non-operator dependent evaluation of the machine vibroacoustic quality. It has to be reliable and effective in identifying faulty products and classifying the defects at the end of the production line, with a confidence level more than 99%. The maximum time required for every test should be less than 2 minutes in order to be acceptable by industry. The prototype test station will embody and rely on two main technological developments.


A prototype test station for house-hold appliance on-line inspection is constructed, capable to give a global evaluation of the vibroacoustic quality of the machine under test, classifying it into two classes, product complying or not to specifications, according to a pass or fail quality control strategy. With the aid of a non-invasive optical sensor network (laser inteferometers) for multi-point vibration measurement the prototype can measure accurately vibrations on the surface of the appliance in a range of frequencies from 0 to 10 kHz, with peak velocity up to 0.2 m/s and resolution less than 1 micron. A diagnostic expert system is capable to identify the typical mechanical defects of house-hold appliances by vibration analysis performed by means of intelligent techniques like fuzzy logic and neural networks.

Status and results

The prototype achieved very high (~98%) classification results. The appliances are mounted easily on the measuring palette and procedure can be accomplished in the predefined 2 minutes lag by setting the appliance to work on the centrifuge mode.

Adaptivity and portability

The system needs new training even if small changes (new materials in parts) occur in the production chain as vibration signatures can be influenced by small relocation of mass distribution of the appliance. Retraining in a new model is a tedious and time consuming task. Cumulating new training data for the data base can be achieved only after a quite large number of tests. Failures in this case have to be artificially produced in order to have a statistically acceptable number of samples. The system cannot easily be relocated and work in new environment without tuning but this tuning refers more to the sensor setup.

More information

MEDEA project, SMT4952016 (search CORDIS page

1997 Stavrakakis G., Pouliezos A., Zervakis M., Anagnostakis E., Tselentis G., "Feature extraction and Neural-Fuzzy Classification of Washing Machines Vibration Signals for On-line Quality Control Purposes", EUFIT '97: European Congress on Intelligent Techniques, ELITE foundation, Aachen, Germany, pp. 1744-1748

technical reports & case studies

executive summaries

scientific papers




search eunite with