When manufacturing hundreds of millions of parts per year, even a microscopic drop in quality assurance accuracy can lead to massive financial losses. For Turvanasta, a leading Finnish manufacturer of winter tire studs, maintaining their reputation as the best supplier in the field means adhering to extremely strict, contractual technical requirements.
Faced with the limitations of legacy quality control systems, Turvanasta partnered with MarshallAI to deploy a high-speed, sub-millimeter defect detection system capable of keeping up with their massive production volume.
The Challenge: High Speeds and Sub-Millimeter Tolerances
Turvanasta’s production environment is the ultimate stress test for machine vision. They operate a broad product portfolio on very fast production lines. The physical products are incredibly small—with stud lengths hovering around 10 to 11 millimeters and weights often under a single gram.
Because the tolerance for errors is practically zero, the existing production machines struggled with a classic quality assurance dilemma:
- High False Reject Rates: The system was unnecessarily rejecting perfectly good production batches, leading to unacceptable levels of material waste and lost revenue.
- Defect Escapes: Conversely, the legacy logic occasionally let critical sub-millimeter defects slip through to the final product.
The AI needed to be capable of instantly inspecting highly reflective metal surfaces across multiple variations, including:
- Aluminium Studs (8-10/2A & 8-11/2A): ~0.90g weight, 10-11mm length, 8mm flange.
- Steel Studs (8-10/2T & 8-11/2T): ~2.00g weight, 10-11mm length, 7.8mm flange.
The Solution: A Three-Phase Deployment
To ensure a seamless transition without disrupting Turvanasta’s massive output, MarshallAI deployed the solution in three agile phases.
Phase 1: Rapid Feasibility Study
Manufacturers often fear that AI requires months of research to prove its viability. MarshallAI ingested image datasets of Turvanasta’s good and defective studs and trained a custom neural network. Within just two weeks, the MarshallAI platform demonstrated a 100% image interpretation success rate on the sub-millimeter defects.
Phase 2: Live Proof of Concept (PoC) at the Edge
With the model validated, the system was moved to the factory floor. MarshallAI installed an industrial camera and a localized edge computing unit directly on the production line. To enable direct hardware actuation, the AI was integrated with Moxa industrial communication modules.
During this three-month PoC, the edge AI inspected parts in real-time at full production speed. The result was immediate: the system drastically reduced the false-reject rate, saving significant amounts of raw material and reducing waste.
Phase 3: Factory-Wide Scaling
Following the resounding success of the PoC, Turvanasta authorized the scaling of the solution. The hardware and software were rolled out in two phases across all active production lines. Rather than treating this as a static software purchase, MarshallAI and Turvanasta established an ongoing “AI-as-a-Service” partnership, ensuring continuous model refinement and development for new stud variations in the coming years.
The Results
By upgrading from traditional machine logic to MarshallAI’s deep learning platform, Turvanasta successfully secured their contractual quality requirements while simultaneously increasing their yield. The project proved that even at extreme production volumes, sub-millimeter defect detection can be fully automated—eliminating unnecessary waste and protecting the manufacturer’s bottom line.
