General Motors

General Motors operates body shops across dozens of assembly plants worldwide, each running high-volume resistance spot welding operations that are critical to vehicle structural integrity. Welding quality in automotive body shops is notoriously difficult to maintain at scale: hundreds of welding guns, each with dozens of process parameters, must stay within tight tolerances across multiple shifts and facilities. Even minor parameter drift — electrode wear, current variation, force inconsistency — can produce weak or incomplete welds that only surface during downstream inspection or, worse, in warranty claims. The cumulative cost of rework, scrapped panels, and production line stoppages across a global manufacturing footprint made welding variability a significant operational liability.

GM partnered with Sight Machine to deploy their manufacturing analytics platform across multiple body shop operations. The platform ingested real-time process data from welding controllers and plant historians, structuring it into a unified data model that allowed cross-plant comparison for the first time. Sight Machine's predictive ML models were trained on historical weld process data to detect parameter drift before it produced defects — correlating variables such as weld current, electrode force, and tip dress cycles with quality outcomes measured downstream. Rather than replacing existing control infrastructure, the platform integrated with GM's existing programmable logic controllers (PLCs) and MES systems, enabling a non-disruptive rollout that expanded progressively across body shops. Process engineers received dashboards surfacing leading indicators of quality risk, shifting quality management from reactive inspection to proactive intervention.

The deployment delivered measurable quality improvements across GM's connected body shop operations:

• 15% reduction in weld defects, directly lowering the rate of non-conforming welds requiring rework or scrap
• Significant reduction in rework time, freeing production capacity previously consumed by defect remediation
• Multi-plant coverage, with the platform now monitoring welding operations across several GM body shops simultaneously

Beyond the headline numbers, the implementation shifted how GM's process engineers interact with welding data — moving from periodic manual reviews to continuous, data-driven monitoring. Cross-plant visibility also enabled teams to identify which facilities maintained best-practice parameters and replicate those settings systematically.

• Connect plants before optimizing them — GM's ability to compare parameters across facilities was essential; single-plant pilots miss systemic patterns.
• Integrate with existing control systems rather than replacing them; non-disruptive deployment accelerates adoption and reduces implementation risk.
• Predictive ML delivers more value than statistical process control alone when weld parameter interactions are complex and high-dimensional.
• Measure leading indicators, not just defect rates — catching parameter drift before it produces scrap is where the real ROI lies.
• Multi-site manufacturers should standardize data models early; inconsistent tagging across plants is the most common barrier to cross-facility analytics.