Case Studies
AI that earned its place on the floor
Problem, approach, and measured result — and how we kept each one validated and audit-ready.
The case studies below are representative and anonymized to protect client confidentiality. They illustrate our typical approach and the kind of outcomes engagements target; figures are indicative ranges, not guarantees, and vary with each operation.
Cutting scrap on a high-cavitation tool with ML process control
Challenge
A Class II molder ran a 32-cavity tool with persistent scrap from short shots and dimensional drift, especially when switching resin lots. Operators chased the process manually, and scrap spiked after every material change.
Approach
- Characterized the baseline with scientific molding and validated existing sensor data quality
- Trained an ML surrogate model mapping process parameters and material state to part quality
- Used multi-objective optimization to balance scrap, cycle time, and dimensional capability
- Deployed parameter recommendations, then closed the loop for lot-to-lot adaptation
Result
- Scrap reduced in the high-teens %
- Reduced dimensional drift across lots
- Faster, calmer material changeovers
Kept Audit-Ready
Model and acceptance criteria documented; changes managed under the site's process-validation framework.
Validated vision inspection that passed a notified-body audit
Challenge
A manufacturer wanted automated vision inspection for cosmetic and dimensional defects on an implantable-device component, but prior attempts stalled because the model couldn't be validated or explained to auditors.
Approach
- Defined intended use, operating range, and acceptance criteria (sensitivity / false-negative limits) up front
- Applied Computer Software Assurance (CSA) — deepest testing on the highest-risk failure modes
- Built explainability into the model so rejections could be understood by a quality engineer
- Produced IQ/OQ/PQ-style validation evidence and a model-governance plan
Result
- Inspection model validated and in production
- Clean notified-body audit of the system
- Lower manual inspection load
Kept Audit-Ready
Full validation package, explainability evidence, and a Predetermined Change Control plan for future retraining.
Targeted predictive maintenance on the assets that actually mattered
Challenge
An automotive molder lost expensive production time to unplanned press and auxiliary-equipment failures, and a prior 'sensors everywhere' pilot had produced data nobody used.
Approach
- Ranked assets by downtime cost and failure predictability — focused only on the high-consequence few
- Built the IIoT data foundation (reliable time-series collection) before any modeling
- Modeled degradation signatures on critical presses and auxiliary equipment
- Set condition thresholds tied to maintenance workflows, not just dashboards
Result
- Fewer unplanned stoppages on targeted assets
- Higher OEE on the affected lines
- A data foundation reused for later AI work
Kept Audit-Ready
Scoped honestly — assets where predictive maintenance didn't pay were left on preventive schedules.
Could your operation be the next one?
Start with a fixed-scope AI Readiness Assessment to see where AI would actually pay off in your plant.
Get Started
Let's talk about your numbers
Every engagement starts with your process and your data. Schedule a consultation, or start with a fixed-scope assessment.