高分机械 research proposal.docx

Research ProposalEarly inference on reliability of automotive components by using past data and technical information1 Background1.1 SignificanceWhen a new product is the result of design or process improvements introduced in its predecessors, then the past failure data and the expert technical knowledge constitute a valuable source of information that can lead to a more accurate reliability estimate of the upgraded product. This proposal introduces a Bayesian procedure to formalize the prior information available about the failure probability of an upgraded automotive component. The elicitation process makes use of the failure data of the past product, the designer information on the effectiveness of planned design/process modifications, information on actual working conditions of the upgraded component and, for outsourced components, technical knowledge on the effect of possible cost reductions. By using the proposed procedure, more accurate estimates of the failure probability can arise. The number of failed items in a future population of vehicles is also predicted to measure the effect of a possible extension of the warranty period. Finally, the proposed procedure was applied to a case study and its feasibility in supporting reliability estimation is illustrated.The automotive manufacturers perform reliability analyses on their new products in order to verify the attainment of a given reliability target at a specified period t0 (e.g., the warranty period or the full life).This aims both to satisfy the expectation soft heir customers and to control there pair/substitution costs during the warranty period. If the target is not achieved, then the product reliability has to be timely improved by identifying the main failure modes and introducing design and/or process modifications capable to remove them. Main reliability evaluations are generally performed at four specific milestones, which occur during the development and after the commercialization of a new car model .In particular: I. At the beginning of the product development process (PDP) to validate the choices made for product and process design: the outcome is the final design release. The main inferential problem in such a phase is the lack of experimental reliability data about the new components. II. During the last phase of PDP, when the final experimental reliability certification is carried out on the basis of component bench tests, experimental design verification tests and product validation tests (including defects arising at the starting of the production process). Since it is often necessary to compress this phase in order to balance all the delays of the previous phases, the main problem is the testing time, immediately followed by the experimental cost; both imply small sample sizes and, consequently, low accuracy of estimates. III. A few months after the commercial launch, when field data are analyzed in order to certify the actual reliability perceived by the customers. Due to the importance for a manufacturer of a new model launch, this information is needed largely before the end of the warranty period. Thus, the main inferential problem is the forecasting in both time and mileage beyond the values reached by most vehicles.IV. During periodic field surveys, where reliability predictions are requested especially in order to check the effectiveness of design modifications introduced as a result of the analyses developed in the previous phase III. If we could wait for the end of the warranty period, we would have complete information based on large samples and no problem at all, but, if we want to advance it, we have to deal with the same forecasting problem in time and in mileage of phase III.1.2 Listing review Information on the failure probability of the past product is generally available under the form of the fraction of failures p0 observed in the population in use during the warranty period t0. By assuming that each failure of a component is assigned to one single part i (i = 1, 2,.) of the component, past data consist of the fraction of failures p0,i experienced by each part i, and the failure probability of the whole component is.In the development phase of a new product, on the basis of the observed failure data and failure modes analyses, the designer attempts to remove the critical failure modes, through design or process modifications. When effective, such modifications will produce a reduction of the failure probability p0,i of the part i of the new product.We assume that the designer is able to predict, for each part, an expected value for the failure probability, say L0,i, that can be achieved only when the planned modification is fully effective. Design and process modifications usually eliminate most of the previous defects, but they can introduce some new ones (that can be detected and eventually eliminated only after the commercialization). Then, the value L0,i may be better considered as the lower limit of the failure probability. Of course, this predicted value refers to the same environmental and operating conditions that have characterized the past product.2 MethodologyThe failure data of the past product refer to the warranty period t0. However, in order to obtain timely estimates of the failure probability of the new product, the reliability team analyzes the failure data of the new product as vehicles are sold and the authorized dealerships perform the repairs. Thus, such warranty data refer to the homogeneous population of a given type of vehicles that use the upgraded component and are observed during the whole warranty period t0 or the early part t t0 of it. The data typically consist in the number of vehicles sold in each month and in the component failures experienced by these vehicles. For each failure, the warranty data base registers both the calendar repair date and the kilometers approximately covered by the vehicle up to the failure. For most components, the driving factor for reliability is mileage x rather than time t.Failure data are usually grouped into mileage bands of convenient width, both to quicken calculations and due to the frequently inaccurate recording of the exact mileage covered by the vehicle up to the failure occurrence. Indeed, when the failure is not catastrophic, the vehicle is often not immediately brought to the authorized dealership after the failure occurrence. Also, in order to analyze warranty data correctly, the number of vehicles with good component has to be distributed in mileage bands, on the basis of the operating time t of each vehicle and ad hoc surveys on vehicle mile age distribution.Since a vehicle is a repairable system, warranty data may include multiple claims from the same vehicle originated by a given component. In the following analysis, however, we assume that no multiple failures are present. The effect of such an assumption should be negligible because multiple failures of a same component usually represent a very small fraction of failures observed in the warranty period.3 Potential ConclusionsThe proposed procedure has showed to be suitable for making inference on the reliability of upgraded automotive components when technical knowledge on the design modifications is actually available or can be asked to the designers and technologists in terms of qualitative factors, even in absence of field data of the upgraded component. This methodology is greatly helpful in all the main predictions, during the product development process as well as later after the product commercialization. The reliability estimate, initially based on few failure data of the upgraded product, can be easily updated as new field data become available to the reliability team, without waiting for the warranty period to be expired by each vehicle.To measure the gain in using the proposed procedure, the reliability of the upgraded product has been estimated under the assumption that no information on the failure probability p0 can be elicited. Then, a vague prior density, given by the product of the non-informative prior density on p0 and the Uniform prior on over the interval is used. We note that the use of the informative prior density on p0 sensibly reduces the uncertainty on the reliability of the upgraded product, both in terms of the failure probability and in terms of the number of items that will fail in the future population, thus supplying the decision maker with a useful tool.4 ScheduleSep. 2014 - Mar. 2015 Systematic literature reviveMar.2015 -May.2015Research on graph partitioning approaching approaches. May.2015-Aug.2015 Code writing and experiments. Aug.2015-Nov.2015 Further refinements. Nov.2015-Dec.2015 Write the thesis.5 DeliverablesI will publish a paper on Chinese core periodicals in December 2015. 6 FeasibilityThe presented case study regards the reliability prediction on a newly revised simple subsystem assembled in a car model already commercialized. All data have been slightly modified to protect proprietary information. This prediction, belonging to the milestone (IV) category, is similar to that of milestone (III) for the whole vehicle. The automotive manufacturer intends to estimate the failure probability of the upgraded version of the outsourced subsystem C both up to the current warranty period of three years and up to an extended warranty period of five years. The upgraded component will be mounted on different car models that will presumably operate under different conditions.7 Reference1Guida, M., Pulcini, G., 2002. Automotive reliability inference based on past data and technical knowledge. Reliability Eng. Syst. Safety 76, 129137. 2Lu, M.W., 1998. Automotive reliability prediction based on early field failure warranty data. Quality Reliability Eng. Inte