Reliability data mining: proposal of a field data processing model

In modern production scenarios, manufacturing companies have the opportunity to collect a lot of data. These may concern for example, production, sales, quality or maintenance of the goods manufactured. The ability to collect data, however, is rarely a conscious and comprehensive data storage activity. The data collection, is not designed for a possible and specific use, but, rather, it is decided in the beginning to collect data and, then, to try to extract some useful information. An appropriate data processing could allow to pass from a disrupted collection of data to a much more useful knowledge. The interest to understand whether some data, collected by a major corporation, could be exploited has led to a study that would allow to understand if it was worthwhile to invest resources in the integrated data collection and creation of a data warehouse. The study was intended to enable us to understand what new knowledge could be extracted from the data, trying to analyse a small sample. The activity of this study was initiated with the intent to explore the possibility of extracting useful information from a data set collected by the company through the development of a data mining model. The analysis was performed on data gathered during the technical assistance in a period of one year, using techniques such as contingency analysis and cluster analysis, developed with a statistical software. After an initial comprehensive analysis of the statistical sample, the data processing activity was carried out in four stages. The first step was the first level contingency analysis, aiming at studying the functioning of the goods produced in the immediacy of the fault. The second step was a cluster analysis, which aimed to define any classes of operation. The third step was a second-level contingency analysis, aiming at verifying the correlation between the fault and the membership to one of the groups previously defined. Finally, we performed a second level cluster analysis, to identify possible classes of machine use, considering, this time, the entire productive life. Despite the relative shortage of data, the project's objective was achieved, confirming the presence of patterns and the possibility of extracting useful knowledge to various business sectors, from design to the identification of faults. The project results have persuaded the company of the high quality of information obtained and the operation of designing an integrated system for data acquisition, storage and extraction of information, following the procedure of mining development, are well advanced.