Issue link: https://iconnect007.uberflip.com/i/1426508
58 DESIGN007 MAGAZINE I NOVEMBER 2021 any other thing you did (or didn't do), your final cost could be impacted. How many times have you been told that, though your plan sounds brilliant, it is much too expensive to produce? How long did it take before you realized this was based on past experience or lack of data in regard to the cost of raw materials or working hours, and that the production could have been cheaper than expected? How many times have you, on the other hand, plunged into what seemed to be the next bonanza, only to discover, too late into the project, that you have based your entire profit- ability prospects on an inaccurate quote which has le you no option but to stop production or go through a new design all over again? is causes you major setbacks in lead-time, not to mention the loss of the capital you have invested. In both cases, you will not make it on time. In a competitive, fast-track market such as ours, that means you lost. Any new design must strive for an accurate cost at the very early stages of its design. For the in-house cost, you must have the accurate data to support immediate costing, while for the outsourced part you need a speedy quote. is information would help you to predict the full system production cost and bring you to the finish line on time. One could say that new designs are destined for inaccuracy simply because they are new, but that is not true. And in a world built on time to market, speed, and customization, companies going through new product introduction (NPI) phases, looking for high-complex- ity board manufacturing, need to stay on top of their game in fear of their prod- ucts becoming irrelevant to the market. A fast quote may become handy in these cases and shorten the lead time by 10–20%. We decided to make use of an automated machine learning system to serve as an infra- structure endlessly collecting data on every item or part number in the production process. Every PCB's design file is uploaded once it is received from the customer to the system, initiating a classification process that, from there on, is automatically boosted with additional relevant data. In the short term, we don't always see the point of gathering all these facts but in the long run, they might be used for solving problems we couldn't have anticipated. Not only that, but the use of this accumulated data (the more the mer- rier) when used by big data systems can be re- evaluated and given a new perspective, simpli- fying complex processes. e system can hand pick the necessary information it requires from all the data to complete its mission. We are not only interested in theoretical engineering data but also in what is practiced on the actual production floor. A database of hundreds of thousands of work orders, failure analysis, the precise production stage at which the error occurred, etc., gives us an indication of homogenous population characterization Figure 1: Our pricing fit model is based on more than 6,000 different part numbers that have completed the full engineering process.