Data Mining Of Failure Cases Based On Traditional Statistical Analysis And Co-word Analysis

By establishing failure database,failure cases are stored in the form of documents,and technical personnel can search and view them according to keywords.However,the abundant information contained in these documents has not been effectively utilized.Co-word analysis is a method to quantify the co-occurrence information of text,widely used in text mining.In this paper,the traditional statistical analysis method and the co-word analysis method are used to analyze the failure cases of Aeronautical Industry and thermal power plant,which are two main fields in failure database of research laboratory.Then the results of the two methods are compared.The structured data including failed components,failure modes and materials in aviation industry related cases are summarized by traditional statistical analysis method.The distribution characteristics of failure components,failure modes and failure materials in aviation industry products are obtained.By introducing the Co-word model into text mining of boiler failure cases,all cases are classified into eight categories:high temperature sulfur corrosion failure at flue side,fly ash wear and dew point corrosion failure of economizer,stress corrosion failure caused by chlorine,thermal fatigue failure of desuperheater and main steam pipeline,failure of welding area of header,water side corrosion failure of water wall,short-term overheating failure caused by scaling or foreign body blockage and long-term overheating failure of superheaters.By comparing the two methods,it is found that the traditional pie chart analysis can only represent the basic data and analyze two indicators at most.Co-word analysis is a multi-dimensional data analysis method,based on the number of simultaneous occurrences of multi-dimensional indicators in a domain text,showing the correlation characteristics between different indicators.