SECMC operates Pakistan's first opencast lignite mine in Block II of the Thar coalfield, Tharparkar, where warehouse and logistics yards support continuous equipment servicing and material handling. Unlike opencast haul roads, the warehouse floor is a confined, high-traffic zone combining routine forklift movement, frequent pedestrian crossing, and limited maneuvering space.
An internal safety audit identified the area as one of the site's most vulnerable man-machine interaction points, driven by blind spots inherent to forklift design, multi-directional operation, and space congestion. This finding formed the basis for selecting the warehouse as the pilot site for AI-based blind spot mitigation.
The deployed Blind Spot Detection (BSD) system comprises four AI-enabled cameras (front, rear, left, and right) feeding a central ECU, a 10.1-in. in-cab monitor, an X-Watch alert display, an audio buzzer, and three-position warning lights. Static blue indicates awareness, flashing red signals an imminent hazard, and bright orange indicates system issues. The AI model is trained specifically to classify pedestrians and forklifts, deliberately excluding general object classes to reduce nuisance alarms. Detection range is configurable up to 13 m, with relay-based outputs available for integration with external equipment.
The project spanned three months from proposal to full deployment, covering vendor evaluation, scope finalisation, procurement, and one week of on-site installation across all four forklifts: two 3 t units, one 5 t unit and one 8 t unit. Two site-specific customisations were made during commissioning.
The detection radius was reduced from 3 m to 2.5 m on the two 3 t forklifts to reduce over-triggering in tighter warehouse aisles. Audio alerts were also reconfigured into the local language to improve operator comprehension and reaction time. The system was not integrated with SECMC's existing CCTV infrastructure and operates as a standalone unit per vehicle.
Three practical challenges emerged during rollout. Ambient dust required ongoing lens cleaning and inspection to sustain camera detection reliability. LED warning lights showed reduced conspicuity in daylight, increasing reliance on the audio buzzer and X-Watch display as primary daytime cues, while the lights retained greater value during low-light shifts. Most significantly, powering the system through the forklift's gear/motion circuit would have voided the OEM warranty.
The system was therefore wired to the main power switch, remaining active whenever the vehicle is powered on. This provided always-on monitoring while preserving the warranty.
Initial operational observations have been encouraging. Operator feedback has been positive, with drivers and floor workers demonstrating improved distance-keeping behavior around active forklifts. The system has also repeatedly flagged pedestrians entering blind-spot zones during routine loading and unloading, confirming the core hazard identified during the safety audit.
Planned next steps include extending BSD coverage to workshop-area forklifts, expanding AI training to cover light transport vehicles (LTVs) and additional site-specific object classes, and introducing tiered detection radii of 1 m, 3 m, and 5 m to support graduated alert severity rather than a binary alarm.
This case study demonstrates that a targeted, audit-driven AI blind spot detection deployment can be implemented across a mixed forklift fleet within a compressed timeline. Practical customisation, including localised alerts and tuned detection zones, improved site-level relevance. The identified challenges of dust, daylight visibility, and OEM warranty constraints also provide transferable lessons for similar deployments in confined industrial material-handling environments.
Author note
Eeman Fatima, Assistant Manager Safety Systems