Manual operations feature prominently in the manufacture of many electrical machines. Even though high-volume electrical machine manufacture is dominated by automation, several manufacturing operations continue to involve manual intervention because of the complexity of such operations makes them heavily reliant on high dexterity manual skills and experience. However, quality can be variable due to human involvement. Currently, in order to maintain a high precision of control and required tolerances of the final machine, inspection is performed at various steps during manufacturing and assembly. Detecting a defect at these end-of-line tests can result in significant wasted time and costs due to rework or scrappage. The solution to this problem lies in in-process monitoring particularly for error prone manual operations. This paper presents a literature review of the state-of-the-art available techniques and limitations in process monitoring within the context of electrical machine manufacturing. To quantify the degree of manual activities in process monitoring within electrical machine manufacture, a structured survey of UK based companies was conducted, identifying specific error prone manual processes to target, and gaps in inspection. The survey identified that a significant proportion of activities in electrical machine manufacture are manual, or semi-automated with manual interventions. However, literature review revealed only a limited research in in-process monitoring of manual operations in this area. Finally, two case studies are presented where case study 1 presents a framework for digitisation of a variety of manual manufacturing tasks, and case study 2 demonstrates real-time capture, modelling and analysis of deformable linear objects in electrical machine manufacturing.

A modified Lattice-Boltzmann method is proposed by considering the Klinkenberg effect and adsorbabilitydesorbability for the purpose of simulating methane gas seepage in fissured coal. The results show that the Klinkenberg effect has a little influence on methane gas seepage in fissured coal, so it can be neglected in engineering computations for simplicity. If both the Klinkenberg effect and the adsorbability-desorbability are considered, the Klinkenberg influence on gas pressure decreases as the Darcy coefficient increases. It is found by gas drainage simulations that near a drainage hole, the effect of adsorption and desorption cannot be neglected, and the location of the drainage hole has a great influence on drainage efficient 𝜆 when the hole is just located at the mid-zone of the coal seam, 𝜆 is 0.691808; when the hole is excursion down to 1.0 m from the mid-zone of coal seam, 𝜆 decreases to 0.668631; when the hole is excursion up or down to 2.0 m from the mid-zone of coal seam, 𝜆 decreases to 0.632917. The simulations supply an effective approach for optimizing the gas drainage hole location.

Monitoring of complex industrial processes can be achieved by obtaining process data by utilising various sensing modalities. The recent emergence of deep learning provides a new routine for processing multi-sensor information. However, the learning ability of shallow neural networks is insufficient, and the data amount required by deep networks is often too large for industrial scenarios. This paper provides a novel deep transfer learning method as a possible solution that offers an advantage of better learning ability of the deep network without the requirement for a large amount of training data. This paper presents how Transformer with self-attention trained from natural language can be transferred to the sensor fusion task. Our proposed method is tested on 3 datasets: condition monitoring of a hydraulic system, bearing, and gearbox dataset. The results show that the Transformer trained from natural language can effectively reduce the required data amount for using deep learning in industrial sensor fusion with high prediction accuracy. The difficult and uncertain artificial feature engineering which requires a large workload can also be eliminated, as the deep networks are able to extract features automatically. In addition, the selfattention mechanism of Transformer aids in the identification of critical sensors, hence the interpretability of deep learning in industrial sensor fusion can be improved.

No abstract

No abstract