Non-Parametric Nevada Format Field Data Analysis

Abstract

Product reliability is ultimately reflected and validated in the field with use of products in real environment. With intensified competition in the market place, the product development and validation cycle becomes shorter. This means the products released to the market might not have been fully validated ideally in the development phase. Meanwhile, new technology makes collecting the data from the field more convenient. The best use of field data is crucial in improving product reliability and satisfying customers. On the other hand, field reliability data can be masked, non-homogeneous, seasonal, mixed with non-reliability factors, and incomplete. They may not necessarily accurately reflect inherent product reliability. There have been many data analysis methods developed over the years. Most of them are parametric methods involving time-to- failure distribution or stochastic process. Good summaries of the challenges and methods in the field data analysis can be seen in [1]–[3]. While the parametric data analysis methods are very useful and remains fundamental, they have some limitations summarized below. (2)The mathematical assumptions for applying these methods often time do not hold in real world. (2)The selection of the analysis method depends on specific application. This makes it hard to provide a unified analysis approach in todays' fast pace office environment which requires more standard and unified analysis and reporting tool leading to quick reaction and timely decision making. (3)Parametric methods are typically mathematically complex. It may not be very intuitive to people who are not familiar with the statistics. Industry practices need the methods easy to understand and explain, more objective with less assumptions and data manipulation, and even easy to visualize. (4)Parametric methods typically have high requirement for data accuracy, which is difficult to meet in reality. The methods that relax this requirement but still provides good value can be very much welcome. (5)Parametric methods typically require more detailed analysis, such as distribution type choice, single or mixed distribution, etc. which makes data analysis automation with big data difficult. There have been efforts in analyzing the field data with non-parametric approach, which can avoid or mitigate many difficulties in the parametric analysis methods though it may not reach the detail of the analysis as the parametric methods do. [3]–[6] are good examples. The field data in many applications exhibit a common format [4], [7], [8] in which for each time period such as a month, certain number of products are released to the market; while certain number of failures occurs. This format of data can be called Nevada format data (table or chart). It applies to both non-repairable systems and repairable systems. Some reliability software package like Weibull $++$ from Reliasoft has the function of the warranty data analysis in Nevada format [7]. In this paper, the effort is made to generalize the Nevada format data analysis methodology and make it a unified and relatively simple tool for industry applications. The analysis methods for some typical uses of Nevada format data are presented. Examples are provided throughout the paper. .