24th International Conference on Production Research (ICPR 2017) ISBN: 978-1-60595-507-0 RELIABILITY PREDICTION OF A MECHANICAL COMPONENT THROUGH ACCELERATED LIFE TESTING A. Regattieri 1 , A. Casto 1 , F. Piana 1 , M.Faccio 2 , E. Ferrari 1 1 Department of Industrial Engineering – University of Bologna, Viale Risorgimento, 2 - Bologna – Italy 2 Department of Management and Engineering - University of Padova, Vicenza, Stradella San Nicola 3, 36100 – Vicenza – Italy Abstract The reliability prediction (i.e. life prediction) of components and subsystems is a crucial issue for the efficiency of production systems. This paper presents a reliability assessment method, applied to a mechanical component (i.e. gear unit) of an automated manufacturing system. Study is based on accelerated life test approach (ALT). In the study authors used General log-linear model (GLL) combined with a Weibull distribution for time to failures. Researchers did different experimental tests in the laboratory of the Department of Industrial Engineer – University of Bologna, to estimate model parameters. Expected life of component was then evaluated using the proposed ALT model. Finally, results are compared with actual time-to-failure data coming from field. Time-to-failure was collected during about 10 years, on about 20 production lines using the same gear unit. The accordance between real data and life prediction is satisfying. Results confirm Accelerated Life Testing (ALT) is an efficient and effective assessment method for lifetime prediction based on short time tests. Keywords: Accelerated life testing, reliability, Life-Stress modeling, Weibull Life Distribution, mechanical component. 1 INTRODUCTION Every company is dependent on some type of asset that keeps the business in business – be it a computer, a centrifuge or a megawatt transformer. In a large enterprise, reducing costs related to asset maintenance, repair and ultimate replacement is at the top of management concerns. Downtime in any network, manufacturing or computer system results not only in high repair costs, but in customer dissatisfaction and lower potential sales [1]. Furthermore, harsher industrial competitiveness in innovation, shorter time of products development and higher machines reliability, implies that strategies for products maturation must be always increasingly efficient [2] [3]. To tackle overall efficiency problem, within manufacturing companies, researchers, engineers and manufacturing are always looking for new and innovative methods to evaluate components and systems reliability [4] [5]. In this field, an interesting and even more promising analyses is Accelerated Life Testing (ALT): is used to force failures to occur in less time by applying increased stresses to the component under test [6]. Results should use in design-for reliability processes to assess or demonstrate component and subsystem reliability, certify components, detect failure modes, compare different manufacturers, and a lot of different studies [1]. Core ALT analyses is the identification of the working-load stress levels of a system (or sub-system or component). When the working-load stresses are known is possible to increase (in laboratory) the same stress, but at an increased level. This just to obtain the component failure, in a shorter time than real. Generally, information from tests at high stress levels of accelerating variables (e.g., vibrations, temperature, humidity, etc.) are extrapolated, through a physically reasonable statistical model (e.g., Eiren, Arrhenius, Inverse Power Law), to obtain estimates of life or long-term performance at normal use conditions [1]. Accelerated conditions allow to reduce testing time and so estimate behavioral characteristics of the product or piece of equipment in normal working conditions [7]. Especially for mechanical components and systems these analyses are useful, because they are designed to operate for a long period of time and in such a case, life testing is a relatively lengthy procedure [8]. After a set of high stress levels tests, is possible estimate the normal life distribution under normal working condition, using an accelerate model. To minimize the statistical error of accelerate model extrapolation, reliability experts have developed numerous accelerated life test plans [9]: aim of the ALT applications is make a model of the real behavior of the component to obtain reliability prediction in a shorter time than real working condition. Most difficult part of the analysis is the life duration extrapolation at standard stress conditions from the accelerated test lifetime. During years, researchers proposed many different approaches both for the accelerate tests design and the following data statistical analyses [10] [11] [12]. From a statistical point of view, the prediction of the reliability of a component deals with the determination of the distribution assumed by its time to failure variable [7]. The most used distributions are Weibull, Exponential [13] and Log-normal. Thanks to its flexibility, the most frequently applied is the Weibull distribution [14] [15]. Weibull distribution, applied to the random variable called x, is based on the pdf function () = ఉ ఏ ቀ ௧ ఏ ቁ ఉଵ  ቀ ೟ ഇ ቁ  Where  is the scale and  is the shape factor. Weibull Distribution is used to supporting ALT analysis by Jung et al. [16] for crankshafts reliability prediction and by Charruau, et al. [11] for reliability prediction of electronic boards in the aeronautic field. In both cases, the unknown parameters, α and β, was estimated by the application of the Maximum Likelihood Method (ML), which is a Point Estimator. ML is an effective, well known and used method. It consists in maximizing the likelihood function, that is, the probability of observing the realization of a given sample, conditioned by the values of estimation parameters. Voiculescu et al. [9] compare the Maximum Likelihood method with Bayesian method, usually used in case of small numbers of data to save costs and time. As wrote by Thairaviam [17], one of the most difficult step and core of ALTs analyses, is identification of the relation between failure times in stressed conditions and the estimation of the failure times in work use condition. In literature, there are different approaches, based on 726