Challenge

To develop an artificial intelligence (AI)-based quality inspection system to introduce a no-fault-forward approach on an overhead conveyor-based production line.

Background

Modern trucks deliver significantly higher horsepower, hauling capacities, and efficiency than their historic predecessors. Key to their technological advancement is their precision build, requiring a high degree of quality control.

The origin of truck-building in Leyland can be traced back to The Lancashire Steam Motor Company, created in 1896 to exploit its original product, a 1.5 tonne capacity steam van. Its first petrol-engined vehicle, nicknamed ‘the Pig’, was produced in 1904, followed a year later by the supply of the first Leyland bus for service in London. In 1987, Leyland Trucks was merged with DAF of the Netherlands to form Leyland DAF, before PACCAR acquired DAF in 1996 and Leyland Trucks in 1998.

Leyland Trucks operates from one of Europe’s most advanced truck assembly facilities, the Leyland Assembly Plant. The company’s future development is focused on the use of leading edge applications of information technology in all aspects of truck design, manufacture, procurement and logistics.

Leyland Trucks worked with the University of Sheffield AMRC North West to develop an AI-based quality inspection system, aimed at spotting manufacturing and assembly defects as they occur, eliminating them before the trucks roll out of the factory gates. This no-fault-forward approach is based on detecting faults quickly, and rectifying them as they move through the production line. Leyland Trucks operate a sophisticated overhead conveyor-based production line, where the time available for detecting faults is limited.

Innovation

The talented team at AMRC North West developed a generic, high-speed machine vision pipeline that can collect video and image frames from multiple camera feeds, and transport them to a powerful AI platform, which then detects components and classifies them into pass/fail categories.

This real-time process requires high levels of data streamlining, including capturing, encoding, and transporting image frames from one location to another, within a few milliseconds. The image frames are then processed by a bespoke Convolutional Neural Network (CNN) AI model, which searches for various elements of interest. Multiple layers of AI classify the detected elements into categories - fault or no fault. The entire process, from detecting a component in an engine to identifying a fault, takes 30 milliseconds using the AMRC North West-developed pipeline.

Result

Leyland Trucks’ former process was manual and labour-intensive, requiring highly experienced human craftsmen to move alongside a chassis suspended from the conveyor to manually identify faults. Human errors and focus mean that the process could be affected by variability, whilst the AI process developed by AMRC North West offers consistent performance, operating efficiently without fatigue.

Comparing human and AI performance in fault detection is complex, as each has its own strengths and weaknesses. When presented with a new challenge, a human will often outperform AI by drawing upon accumulated knowledge, reasoning and intuition; if a human worker spots a fault on a fuel pipe assembly, they draw on experience to reach a conclusion, whereas an AI’s knowledge is limited to its trained dataset.

Conversely, while a human’s peak accuracy may be higher in certain scenarios, it is likely to vary between individuals, and deteriorate with extensive repetition due to physical and mental fatigue. AI’s performance, however, remains consistent.

Impact

The AI model can count and report the number of specific components within seconds, instantly identifying any missing parts, and is expected to considerably reduce the incidence of faults progressing further along Leyland’s production line.

The system has been trained to detect clips and clamps, with accuracy levels of 86 per cent and 70 per cent, respectively, further classifying them as fault or not with accuracy levels of 70 per cent and 75 per cent, respectively.

Vehicle quality is paramount in the contemporary manufacturing industry - a missed fault on a production line can lead to significant rework during final testing, which can be costly and cause reputational damage if a vehicle is released with a fault. The Leyland Trucks’ quality assurance team has developed a comprehensive list of checkpoints containing thousands of line items.

Factories are harsh environments, with high noise levels, making inspection difficult with dark colours and metallic reflections. Various sensing techniques have been employed in the AI system, including multiple lighting and exposure settings, and optical filters to enhance sensitivity.

In the future, the AMRC North West project team aims to use 3D laser imaging to improve clarity, so that even minute faults which may be easily missed by the human eye can be detected.