Haul truck tyres are a major operating cost and reliability constraint in iron ore open pit mining. Tyre life is driven by many interacting factors including payload, road condition, and tyre construction; no single measurement captures them in combination. Tyre Pressure and Temperature Monitoring Systems (TPMS) provide useful general health indicators but rely on static thresholds and offer limited insight into internal injuries that progress to tyre fires or full vehicle loss. This paper presents mid-trial outcomes from an Autonomous Inspection System (AIS) deployed at a Pilbara iron ore mine for a six-month trial. Results reported here cover the first three months, January to March 2026; the trial is ongoing at the time of writing. The platform uses solar-powered side-of-road skids equipped with FLIR thermal imaging and computer vision to scan ultra-class haul truck tyres without slowing the fleet. A continuously trained model identifies tread injuries; a static Trigger Action Response Plan (TARP) matrix translates injury dimensions and temperatures into operator-actionable classifications. The system has been in continuous operation across mining sites in Australia, North America, and South America since February 2020, with approximately five million haul truck scans logged to date. Across the three months reported here, the site’s third-party tyre management contractor recorded 36 early life component failures. The AIS auto-detected 28 of these (78 per cent), with three further failures identified manually from system imagery. Average lead time from first detection to removal was 8.8 days; eight detections provided 14 days of lead time or more, enabling intervention within the planned fortnightly maintenance window. A rock cut tread separation on a rear-axle position progressed through HIGH and EXTREME TARP classifications over a sustained multi-day alerting window, and was removed in a vented separation condition with nine days of lead time from first detection. The case illustrates the prediction-to intervention workflow the system enables. The paper presents the failure-mode taxonomy targeted by the system, the validation chain against independent post-removal reports, observed limitations, and continuous-improvement work including front-axle damage detection released in April 2026.