Developments in Advanced Composites Skills Training and Manufacturing through Virtual Reality and an Artificially Intelligent Layup Agent

Abstract

Despite being the dominant manufacturing route, hand layup of composite materials is still poorly understood. Until recently, only very little documented knowledge has existed on how the manual portion of the hand layup process was actually carried out. On a simple level, hand layup is a process whereby flat sheets of mainly pre-impregnated woven clothes (prepreg) are deformed into complex shapes using hands and support tools. It is labour intensive and hence can be prone to large amounts of variation. A large proportion of these variations can be attributed to the operator skill, layup instructions, geometry, and material. A potential workforce shortage, with the necessary skills to undertake the layup tasks, has been identified as one of the major concerns in the near term. Aligned to this is an existing and immediate need for re-skilling and/or up-skilling of the current workforce, as well as standardization of training delivery. Work presented in this thesis has attempted to address these issues by exploiting Virtual Reality (VR) technology, the concept of Gamification, and artificial intelligence techniques. A feasibility study on the use of VR technology and gamification on composites skills/knowledge transfer was carried out on the layup of a flat carbon fibre composite panel using unidirectional prepreg material in a typical composite clean room environment. The VR system used included a smartphone based head mounted display, hand held game pad as the input device and a headphone, which outputs audio instructions to the user. The conclusion from the evaluation of the training aid was that the designed training aid, in its current form is an effective knowledge capture/transfer tool, but so far has limited use in physical skill transfer due to the use of a hand held input device. As a knowledge capture/transfer tool, it showed potential in the reduction of the learning curve of novice laminators. It was also identified that this concept can easily be adapted to layup of more complex parts with the use of appropriate alternative hardware that would enable the transfer of physical skills via hand tracking. A second VR training simulation was designed for a 3D mould shape. The VR system used for this simulation consisted of an Oculus Rift Development Kit 2 and a Leap Motion Controller as a hand tracker. Nineteen distinct actions were identified as required to complete the layup of the chosen mould through video analysis of expert laminators laying up the chosen mould. Relevant experiments were carried out to obtain a preliminary understanding on the suitability of VR technology as a platform for hand lamination training. Overall feedback received from the test candidates suggested that the VR training aid in its current form might not be suitable as a standalone platform for hand lamination training. However, experimental results did not fully agree with the opinions of the test candidates, since test candidates who received VR training only, performed better than the test candidates who received one to one training during their layup trials. Lack of haptic feedback was identified as a major bottleneck in using such VR systems to train laminators, especially when complex mould shapes that require in plane shear to be applied to prepreg material to complete the layup are in consideration. Following on from the two VR related case studies, inability to predict the sequence of actions required to layup a given mould shape was identified as a major issue in designing and potentially automating the generation of such VR training simulations. Previous attempts made at using outputs from existing drape simulators to provide layup instructions were reviewed. The conclusion here was that such previous attempts have failed to provide sufficient information required for shop-floor operatives to carry out layup operations. An artificially intelligent (AI) layup agent capable of determining the ideal layup sequence for a given mould shape based on deep reinforcement learning was proposed as a solution to this issue. Three distinct case studies were carried out to explore working mechanics and limitations of the proposed AI agent. Results from the case studies suggested that, the AI agent in its current form is capable of determining the ideal layup sequence for a given mould shape, if the complexity of the mould shape is not beyond the current capabilities of the AI agent. Factors that limit the current AI agent from being able to handle mould geometries that are more complex have been identified and possible solutions to these limitations have been proposed as future work. The work presented in this thesis has laid down the foundation and introduced new frameworks and use cases for emerging technologies in solving key issues around the hand layup process of prepreg. Both case studies carried out related to VR training for prepreg layup showed an increase in accuracy of layup >20% when compared with traditional training methods. This is a clear illustration that VR training is bound to play a key role in training hand laminators in the near future. In addition, the AI layup agent introduced in this thesis has opened up a completely new research avenue for the development of future drape simulators that will be capable of providing optimized and numerically validated instructions for hand laminators and the design of such VR training simulations.

Date of Award	23 Jun 2020
Original language	English
Awarding Institution	The University of Bristol
Supervisor	Carwyn Ward (Supervisor) & Kevin Potter (Supervisor)