Service Communautaire d'Information sur la Recherche et le Développement - CORDIS

FP6

THERMOFIT Résumé de rapport

Project ID: 513182
Financé au titre de: FP6-SME
Pays: Denmark

Final Report Summary - THERMOFIT (A thermal imaging based part inspection tool to enhance the competitiveness of the large group of SMEs within the EU injection moulding Industry)

The overall objective of the project THERMOFIT was to develop a new low cost thermal imaging based part inspection system that allows injection-moulders to rapidly identify parts that are likely to develop failures, as a consequence of cooling or future expensive reprocessing.

The overall scientific project objectives were to:
- enhance the scientific understanding of the influence of part geometry and material on the temperature profile on polymer components as they are removed from the mould;
- enhance the scientific understanding of the relationship between variations in component surface temperature profile during a production run, and the part quality, both when it has cooled and when it is reheated during post-mould processing.

The technical work has been spread over the tasks in Work packages 1, 2, 3 and 4:
- WP 1: Enhanced scientific understanding of polymer thermal profiles after demoulding
- WP 2: Fault inspection module
- WP 3: Second order part inspection module
- WP 4: Integration and industrial trials.

The kick-off meeting was held at Pera Denmark on 13 January 2005, where the project goals, work plan and initial actions were successfully presented and agreed upon. During the duration of the project meetings were held approximately every three months hosted by various partners reviewing and carrying out trials for their relevant tasks.

One result of first meetings was that the inspection system should be applicable to any part, this being regardless of fault, geometry and material, this posing a challenge regarding the software because no static variables can be used, as these might not work on certain faults, geometries or materials. Hence, all variables should be dynamic, meaning they will be calculated based on the given dataset. For each type of fault THERMOFIT will need to detect, a specific module will have to be developed.

The user interface should be very simple and straightforward, as the end-users are likely to have little or no experience regarding the setup of software or vision systems, and the solution of the problems will therefore be created requiring little or no user setup.

The software setup should also be very simple and fast to configure and use, as software with a complex configuration might not be used. If the machine operator does not understand what he is doing or finds it too time consuming compared to the traditional 'trial and error' approach, the operator is likely not to use the software.

Furthermore, the software should be fast to setup and require as little training as possible, because a system requiring thousands of parts would provide little or no benefit to the average end user running small batches. To sum up the general requirements were for the system to be:
1. adaptable - no static variables concerning part and fault size, geometry and material;
2. simple - limited user configuration;
3. fast - fast training by a reduced amount of features;
4. accurate - high rate of detection;
5. robust - applicable to an industrial environment.

Due to the high adaptability and the proven track record of neural networks within classification and pattern recognition, a solution using neural networks for the fault identification module has been chosen.

Tests have been performed on different material: ABS, PP and PA6; and different type of moulds:
- tests performed on GAIM parts (U-part, spin part) to produce incomplete filling part;
- tests performed on solid part (U-part, Porch part, Micotron mould, Gimplast and Micotron mould) to produced warpage and marks.

Currently, the accuracy of the system, depending on the test data set ranges from 98 % to 100 %. The majority of the performed tests yield very good results. The faulty with no access to the physical parts its extremely difficult to determine why the system detects bad parts as good, the injection moulding process is very difficult to micro manage and small deviations occur between shots; hence, some parts may be borderline good and hence the system detects it.

Informations connexes

Reported by

MICOTRON A/S
6, Lontoft
DK-6400 SOENDERBORG
Denmark
See on map