Avaliação da confiabilidade de bombas centrífugas utilizando dados de acionamento de garantia como falha em campo

Abstract

This paper presents the conduct of a product reliability assessment, using warranty drive data as a basis for analyzing failures in the field. Reliability is the probability that a product will operate fault-free for a specific period of time. Considering that brand image and product credibility are linked to customers' perception of quality, understanding product reliability becomes crucial to remaining competitive in the market. Many manufacturers, occasionally limited by resources, launch products without understanding their reliability, resulting in high repair costs due to warranty returns. In this context, warranty drive data can provide a valuable and low-cost opportunity to understand the reliability of products on the market and apply the concept of continuous improvement, as it offers a realistic view of product operation and failures in the field. The research focuses on assessing the reliability of hydraulic centrifugal pumps, manufactured by a leisure equipment industry with more than 50 years on the market. The aim was to statistically verify product reliability, which has never been assessed before, using field failure data collected by the Customer Service Department (SAC) through warranty claims, thus demonstrating a free and timely method, especially for manufacturers who have products in this context. The Statistical Theory of Reliability was applied, performing a probabilistic modeling of the data up to the first failure, considering a single failure mode (due to limitations of the data used). Fitting them to the Weibull distribution model, using Maximum Likelihood estimation and the Anderson-Darling fit quality test, the aid and demonstration of the use of computational resources with MINITAB® software (closed source) was carried out. The same analysis was also demonstrated using RStudio software (open source) in order to present an accessible resource. Due to the difficulties in adjusting the field failure data derived from the warranty activation (which is of an uncertain and censored nature), the results of the parametric Weibull model were compared with the results obtained by the non-parametric Kaplan-Meier model, generated with the same software. Tools such as Failure Mode and Effect Analysis (FMEA) and Fault Tree Analysis (FTA) were also used to analyze failure causes and effects, complementing the assessment and highlighting possible opportunities for organizing failure data by failure mode. The results obtained using the method showed reliability values of approximately 40% halfway through the warranty period; this result does not seem to accurately represent the reality of the product being assessed and may have been influenced by the uncertain and censored nature of the data derived from the warranty claim, and by the inaccuracies observed in the data supplied by the manufacturer. These results suggest that, in order to take advantage of the benefits of applying the method, it is important to consider the uncertainty and censorship characteristics of the field failure data obtained from the warranty activation records and minimize them in order to obtain more accurate and coherent results. In this context, studies to improve the method are suggested and encouraged, in view of its solid benefits (representing failures in a real scenario, without controlled operating variables, at zero cost) and social impact, which mainly help small and medium-sized companies/industries/manufacturers to implement a reliability-oriented culture.