By: Bala Deshpande, Co-Chair, Predictive Analytics World Manufacturing
In anticipation of his upcoming conference presentation, Predictive Analytics Solution Template for Early Prediction of Assembly Line Failures, at Predictive Analytics World Manufacturing Chicago, June 19-22, 2017, we interviewed George Iordanescu, Data Scientist at Microsoft. View the Q-and-A below for a glimpse of what’s in store at the PAW Manufacturing conference.
Q: What are the challenges in translating the lessons of predictive analytics from other verticals into manufacturing?
A: Rather than challenges, from the Data Science perspective I would say there are actually exciting opportunities. The Cortana Intelligence solution How-to Guide I am presenting showcases an important aspect that is specific to assembly line failures data: We provide a generic Advanced Analytics solution that uses Machine Learning to predict failures before they happen. Early prediction of future failures allows for less expensive repairs or even discarding, which are usually more cost efficient than going through recall and warranty cost.
Q: In your work with predictive analytics, what behavior do your models predict?
A: A key point is that customers’ pain points – the failures- and their subject matter expertise are actually the real goldmine. By specifically looking at returns and functional failures at the end of assembly line and combining these with domain knowledge and root cause analysis we provide a generic advanced analytics solution with a modular design that encapsulates main processing steps and leverages test systems already in place. So no new equipment is necessary, since we use machine learning to detect subtle patterns in test systems measurements that pass the regular quality checks but are still indicative of a future failure.
Q: How does predictive analytics deliver value at your organization? What is one specific way in which it actively drives decisions?
A: A good scenario is the OEMs who use contract manufacturers (Apple-Foxconn, MSFT-Jabil), and want to minimize post sale service and warranty costs: Use ML to build models that use test/shop floor data (that belongs to OEM and or CMs) to predict before shipping, field return and repairs that may happen months or years after the device is shipped. This enables predicting future failures while the device is still in the early manufacturing line stages or is already assembled but is not yet shipped, so that fixing or even discarding it may be cheaper than going through recall and warranty costs.
Q: Can you describe a successful result, such as the predictive lift (or accuracy) of your model or the ROI of an analytics initiative?
A: A good example is Jabil where they are not only predicting that a potential issue or failure could occur, but by having visual insight into all levels of production and operations, the company also now knows why the failure was predicted. That it is actionable information Jabil can use to avoid a costly loss, while shortening product lead times and delivering superior quality.
Q: What surprising discovery have you unearthed in your data?
A: I think it is surprising and very empowering to see that fundamental information comes from the client: Failure data, domain knowledge, and test systems already in place. This setup happens to be a natural example of a Machine Learning concept called Gradient Boosting, where an imperfect (weak) model (i.e., the existing quality check system) is extended and improved by focusing on its errors, i.e., the hard cases given by the assembly line failures.
Q: Sneak preview: Please tell us a take-away that you will provide during your talk at Predictive Analytics World for Manufacturing.
A: We show how a solution can be implemented and deployed fast in the cloud, using the flexible on-line Microsoft Azure platform that decouples infrastructure components (data ingestion, storage, data movement, visualization) from analytics engine that supports modern data science languages like R and Python. The solution modeling component can thus be retrained as needed and be implemented using high performance Azure Machine Learning algorithms, or open source (R/Python) libraries, or from a third-party solution vendor.
Don't miss George’s conference presentation, Predictive Analytics Solution Template for Early Prediction of Assembly Line Failures on Tuesday, June 2017 at 10:30 to 11:15 am at Predictive Analytics World Manufacturing Chicago. Click here to register to attend.
By: Bala Deshpande, Founder, Simafore and Conference Co-Chair of Predictive Analytics World Manufacturing