Automated, scalable robotic NDT inspections of unknown free form parts

Student thesis: Doctoral Thesis


Manufacturing techniques have evolved around requirements of the consumer in the context of competition, with quality assurance central to the 20th century paradigm of consumer spending on high value goods, the mainstay of Non-Destructive Testing (NDT) during this period. Prioritisation of end-use reliability lead to a plethora of techniques such as subtractive manufacturing in which material and energy waste were given attention proportional to their marginal costs at the time. Mass-produced goods in this value bracket were often guaranteed to have homogeneity and so automating their inspection with repeatable, accurate, yet dull robotic platforms provided a natural extension and boon to inspections. In recent years, increasing direct and external costs in energy production and end-of-life waste have exhibited themselves as an increasing focus in government and industry on reducing greenhouse emissions through novel manufacturing methods and closing product life-cycles by way of remanufacturing. Lightweight moulded composites and pre-used parts present a break from the repeatable product archetype that robotic NDT is traditionally suited to, requiring adaptable inspection processes to scan hosts of parts with minor but geometry-deforming defects. Currently, path planning for NDT inspections require either digital representations to create and simulate inspection routines, or for a lengthy operator driven jogging procedure that can present bottlenecks to a scanning process. Repeatably manufactured parts in prior production processes were either provided with a digital model as an accurate template for manufacturing, or were sufficiently similar that a jogged path could be applied to every part produced. Digital path planning on moulded and remanufactured parts is not possible in the same way, requiring extensive metrological inspection prior to planning, with jogged planning completely unfeasible. Separately, the demand for online path planning for mobile robotic arm platforms has seen large growth recent years, driven by a desire for remote scanning in hazardous environments and to increase the inspection through-put of green technology such as wind turbines. In these conditions, a digital representation of the part is not necessarily available or can entail a lengthy digital-world to real-world calibration procedure for path planning to commence. Post inspection, while industrial robotic arms can generally reconstruct NDT data to sub-mm accuracy, a mobile base with poor or no odometric data canentail the use of high-cost, fixed volume metrological equipment to reconstruct data to the same accuracy in order to enable an operator to properly validate, repair, or sentence the part. By taking the novel perspective of minimising the number of mathematical perspectives necessary for each sub-problem, this thesis investigates the minimal quantity of data necessary to profile and inspect an unknown free-form part autonomously and independently of additional equipment, and in the subsequent data reconstruction when multiple scans are taken. Of particular interest is the dry-dock scanning of RNLI Severn class lifeboat hulls, the targeted industrial use case scenario. The output is a novel autonomous path planning system using low-cost RGB/D cameras, laser line sensors and force/torque control that reduces the path planning and deployment time from up to a month down to several minutes. This thesis further provides a novel approach to data stitching with visual reference markers, demonstrating an optimal accuracy in the order of mm. The result is an easy to use process utilising Python software that requires no prior information from a human operator. The process workflow from part placement to part reconstruction is represented in Fig: 1. It is capable of quickly generating and then deploying scan paths for fixed or mobile robotic-arm platforms, and of accurate reconstructing of parts in the latter case.
Date of Award21 Sept 2023
Original languageEnglish
Awarding Institution
  • University Of Strathclyde
SponsorsEPSRC (Engineering and Physical Sciences Research Council)
SupervisorGareth Pierce (Supervisor) & Anthony Gachagan (Supervisor)

Cite this