Our approach to 'Predictive Quality' via Machine Learning and Tooling Data
Updated: Jun 17
Part defects and product failures -- with all the associated costs, brand damage, and recalls -- are the bane of manufacturers. ‘Quality’ is important enough to warrant its own QA/QC specialists, but most defects aren’t discovered until late in the process via quality checks. In the worst case, faulty products will slip through and end up in the hands of customers. Defects should be detected sooner; there’s room for augmenting QA/QC processes, for optimizing costs and process times. To be on the cutting edge, companies need to predict the quality of their parts as soon as they’re manufactured.
eMoldino’s solution plans on providing this ‘predictive quality’. The proposed solution is a machine-learning based system that predicts whether a part is defective, judging by the cycle time, temperature, and pressure of the tooling (or mold, die, press, etc.) at the time the part was created.
This is possible by aggregating, over time, ‘tooling condition’ data, quality data, and parts-returned data. Afterwards, the machine learning system will correlate quality with specific tooling conditions and thus predict part quality. This capability will augment overall QA and QC processes and be invaluable for OEMs and suppliers alike.
Read on to understand the logic behind our predictive solution
"The [solution] will estimate a part's quality by checking the conditions of the tooling at the time the part was produced."
What’s the Relationship between Quality and Tooling Conditions?
At its core, the solution works under a simple assumption: the conditions of the tooling influences the quality of the product.
Even minute variances in tooling conditions (i.e. cycle times, temperature, pressure) will affect the quality of parts and cause defects. Normally, factory operators will try to control these conditions by operating toolings and equipment optimally and consistently. The problem is that, despite the operator’s best efforts, the conditions of a tooling will change throughout production. Cycle times can always vary by a few or half-seconds, while temperature and pressure will shift depending on the stage of production. The superior solution is then to monitor these shifting tooling conditions and correlate them with part quality -- all in real-time.
How Our ‘Predictive Quality’ Works
Thus, the first part of our all-in-one solution is to track tooling conditions: cycle times, temperature, and (in the near future) pressure. Whenever the solution detects a tooling operating outside of optimal (or predetermined) cycle time, temperature, or pressure parameters, it will automatically alert the appropriate end-users.
The second part of the solution involves machine learning. Over time, the system will accumulate a history of tooling condition data. Simultaneously, the system will be fed defect parts and failure rates recorded during quality checks, as well as parts-returned feedback. The machine learning algorithm will continue to churn these data-sets for months; eventually, it will learn to associate certain tooling conditions with either high- or low-quality output.
This means, moving forward, the system will estimate a part's quality by checking the conditions of the tooling at the time the part was produced. For example, if the algorithm predicts that a part will be faulty because it was made when the tooling was too hot, it would flag that part and report it to the end-user.
And of course -- the longer the solution runs, the more data the algorithm gets to consume, the more patterns it will identify, and the more accurate its predictions will be. Consequently, the longer our clients use our solution, the more value they will get from it.
What Does This Mean For You, the Manufacturer?
Condition-tracking your toolings, molds, and dies can predict the quality of a product before having to physically quality-check it. This means less work and more streamlined management for QA specialists and project managers alike.
Our solution’s ‘quality’ metric and related KPIs will be invaluable for OEMs and suppliers. Employees from all points on the value chain will be able to review the quality of specific products on a shared, real-time platform. With the system, managers can assess quality before they have to conduct QC tests, and even augment existing QA processes, saving time and overall costs.
Additionally, our platform will provide OEMs with an auditing trail. This will let them see all the potentially-defective parts that were produced and see which of them were put out to the market. In such a way, questionable parts may be located, returned, and reassessed with greater ease.
There are very few solutions in the market that provide ‘predictive quality’ metrics, and even fewer that use real-time data to formulate them. Our predictive paradigm is possible because we focus, uniquely, on the tooling level (molds, dies, presses, and so on). Discuss the potentials of this exciting first-to-market solution by arranging a conference call with us, with no obligations. Feel free to ask about our solution’s data-reporting analytics, and see which of them will directly impact your professional area.