No Problem Found Framework Based on Analytics and Machine Learning

Summary & ConclusionsDisruptive events from field returns are difficult to analyze at best and can result in a "no problem found" (NPF) classification. The percentage of NPF is very high, ranging from 40% to 70% of customer returned products [1]–[3]. The difficulty is further compounded when the customers' environment cannot be reproduced at the manufacturer diagnostics testing facility, whereby uncharacterized NPF results in incomplete information and risks in future product designs.The analytics framework assists in looking for areas that would lead the reliability engineer to the issue of component vulnerability, which should minimize the complex process of elimination. A methodology of correlations and query lookups will be much faster compared to iterative supplier interactions, which could take weeks. The framework is not going to solve the problem. The knowledge base and big data can collate information, with the intent of cutting time, by pinpointing areas that would prioritize the diagnosis path, rather than solve every possible scenario.Applying analytics for customer returns works best when problems can be found but leaves "no problem found" as unidentified with no ability to diagnose, predict and prescribe as shown in Fig. 1. In order to address the unidentified analytics, we must correlate components with other known products, customer usage, and supplier provided information. This methodology requires the big data approach in order to capture as much data as possible from the manufacturer, supplier, and customer environments. Correlation, pattern discovery, and trends are built on using machine learning techniques. The value in big data and machine learning is the speed in which the reliability engineer can correlate this data.This framework integrates several modeling approaches which addresses the unidentified areas due to NPF. These approaches combines business analytics process, big data, bio-inspired behavioral modelling and machine learning artificial intelligence (ML/AI) techniques to discover new relationships. Big data typically means more data is better. Traditionally, the manufacturer only tracked component data on the board in order to analyze NPFs from customer returns.A knowledge base allows for integration of existing big data information from various environments relating to the board. These sources of information cover the scope from supplier, manufacturer and customer. NPF returns are tracked using bio-inspired behavioral models. The business analytics allows manufacturers to be proactive and predictive for NPF. Self-organized maps are used as an ML/AI technique to discover new relationships.In conclusion, the business analytics framework and architecture is improved by a bio-inspired model which allows business analytics to become proactive and predictive. The big data methodology allows us to expand the parameters for the supplier, manufacturer and customer in order to make better decisions. It also helps focus attention on analysis and consider what is significant and not significant. Furthermore, it helps reduce sustainment cost across multiple product lines. The machine learning discovers new failure mode relationships between the three mission profile environments. This enables pinpointing of diagnostic paths for the No Problem Found (NPF).