A non-conventional quality control system to detect surface faults in mechanical front seals

https://doi.org/10.1016/j.engappai.2007.11.007Get rights and content

Abstract

The Just in Time and the Total Quality policies have remarkably touched every field of modern industrial production. This context prompts companies to dedicate most of their efforts on researching and developing automatic systems of quality control to obtain the elevated standards of quality nowadays demanded by the market at every level of production. In fact the quantity of the exemplars allowed, which are not up to sample, is measured in parts per million in many sectors of production.

Many methodologies have been proposed to yield high quality in the industrial production lines, in order to provide surface examination and classification.

This paper describes an alternative system for surface analysis based on artificial neural networks (ANNs), developed in collaboration with the Italian manufacturer “Meccanotecnica Umbra S.p.A”. This system was implemented and tested in order to examine three particular surfaces of mechanical seals achieving good results in comparison with the deterministic system already implemented.

Section snippets

Introduction and methodology

This study aims to develop a quality control system to detect some particular faults, currently not recognized by the deterministic system already implemented by the Italian manufacturer “Meccanotecnica Umbra S.p.A”, that affect mechanical frontal seals. Consequently success relative to the detection of these faults means a direct improvement as compared with the methodology already implemented.

Therefore, the analysis is focused on the three particular faults currently not recognized and listed

Input parameters

As can be noticed from the images shown in Fig. 2, the surface area to analyze is dimensioned to be comparable with the dimensions of common faults, so that the surface imperfection meaningfully affects the statistical parameters extrapolated from such a zone.

Therefore, the images are acquired as RGB (1600×1200 pixels) and then converted to gray scale. On the basis of these data, it was noted that the presence of typical faults on the surfaces taken in consideration causes a sensible variation

Container bottom surface analysis

In this case the training set collects 20 images of good exemplars and 20 images of seals with scratches on the container's bottom surface. The unsuitable seals were placed in front of the camera so that the scratches were all along the analysis area. The testing set collects 20 images of good seals and 10 images of scratched ones. This set was used to test the performances of the trained neural system.

As already said, the input array collects 25 elements per image; consequently 25 neurons make

Container lateral surface analysis

The system implemented to analyze the bottom metal surface of the container was adapted to control the quality of the lateral one. As already said, it is necessary to identify the lack of seal on this surface. Even in this case a training set and a testing one were collected.

In reference to the ANN architecture of Fig. A.1 in Appendix A, the best grade of training convergence was reached for the configuration 25-30-20-10-1. After 30 000 epochs and a training time of 15 min, SSE and RMS values

Carbon surface analysis

Finally, the system was adapted to identify imperfections on the carbon surface of the rings that are the internal elements of mechanical seals. A training set of 40 (20 OK–20 NOK) images and a testing one constituted of 27 (18 OK–9 NOK) exemplars were collected.

In this case the configuration, characterized by a number of neurons in the input, inner and output layers of 25-30-30-10-1, guarantees the best grade of training convergence after 30 000 training epochs (convergence time of 15 min). The

System performance adding artificial noise to input images

As discussed in Section 1, all the images object of this study were acquired with a common camera and, therefore, already characterized by an intrinsic noise in comparison with the ones which in the implementation phase of the quality control system will be acquired by means of a suitable vision system.

Anyway, in order to carry out a sensitivity analysis of system according to the quality of the acquired images, it is interesting to verify the performances in the case of additional noise

Conclusions

Applied to all classes of analyzed defects, the implemented method achieves high level of accuracy for not-high quality input images and also in the case of supplementary low-level noise introduction. In fact, as seen in the previous paragraphs, the neural classifier, applied to the three defects (lack of seal, scratches on the metal surface and carbon ring defects), is able to separate 100% of faulty elements from the good ones. One more important aspect to underline is that, once the system

References (10)

There are more references available in the full text version of this article.

Cited by (7)

  • Knowledge-driven intelligent quality problem-solving system in the automotive industry

    2018, Advanced Engineering Informatics
    Citation Excerpt :

    Besides, the scope of quality management covered by these studies is too broad due to the lack of in-depth study of the detailed activities of quality management. At the same time, some scholars have paid close attention to performing quality inspections to improve or predict quality through process flow parameters and on applying intelligent technologies to quality management tools [12–17]. However, few scholars have studied the application of intelligent technologies in quality problem-solving activities, such as causal analysis and the formulation of measures.

  • E-quality control: A support vector machines approach

    2016, Journal of Computational Design and Engineering
    Citation Excerpt :

    Multi-class support vector machines were used for the fault diagnostics of roller bearing using kernel based neighborhood score multi-class support vector machine, where it was shown the multi-class SVM was effective in diagnosing the fault conditions and the results were comparable with binary SVM [14]. Artificial neural networks were used for addressing quality control issue as a non-conventional way to detect surface faults in mechanical front seals, which achieved good results in comparison with the deterministic system which was already implemented [15]. Fuzzy association rules were used to develop an intelligent quality management approach with the research providing a generic methodology with knowledge discovery and the cooperative ability for monitoring the process effectively and efficiently [16].

  • Detection and classification of surface defects of gun barrels using computer vision and machine learning

    2015, Measurement: Journal of the International Measurement Confederation
    Citation Excerpt :

    In this paper, a framework is developed to detect and classify the defects appearing on the surface of gun barrels using machine vision approach. Visual inspection of surface has been carried out on a vast variety of products in past few years like front seals [3] and PU packing [4]. Various segmentation methods for detecting defects have been proposed and implemented successfully [5,6].

  • Visual inspection of products with geometrical quality characteristics of known tolerances

    2010, Ain Shams Engineering Journal
    Citation Excerpt :

    Their approach is based on Hotelling statistics and image analysis. Other examples includes [1,3,27,29]. The decision of all developed approaches is either to accept or reject the product without determining the source of defect(s).

View all citing articles on Scopus
View full text