Real-Time Deep Learning for Leaks Detection and Demands Prediction of Compressed Medical Gas Systems

The increasing pressure on the NHS has accelerated the digital transformation of healthcare technology by building smart hospitals as net zero carbon buildings with more sustainable and energy-efficient operation of structures. This has reinforced the need for technology enablement to support large scale transformation and models to adapt rapidly to support real-time remote monitoring and predictive control of healthcare estates such as compressed medical gas systems (CMGSs), where the combinations of reliability, energy inefficiency, energy loss by leakages, air quality and plant downtime are considered as the key risks. However, due to the complexity in different NHS contexts, to reliably predict demands for optimal control and detect leaks developed over time in compressed air systems is still a challenge. To suit the needs of healthcare industry, this paper proposed a novel AI-based data-driven system to make production processes more efficient and adaptive using artificial intelligence (AI) through big data analytics over Internet of things (IoT). By developing deep wavelet neural networks (DWNNs) with structure simplified and optimised for fast learning and using tight frame Mexican hat (Gaussian) wavelets with sparsity as activation functions for better generalisation to process streaming data in time and frequency domain, the real-time AI system designed can achieve optimum performance of medical air and vacuum plants with power efficiency optimised by predictive control. Through qualitative and quantitative data analytics, the real-time deep learning allows to provide valuable insights, track everything that is relevant to operations, and predict and optimise system performance. Experimental results demonstrate that optimal operations of CMGSs can thus be achieved by the real-time AI, where remotely collected data communication, processing and analytics are provided by the designed controller, communicator and distributed cloud, such that by real-time deep learning, both knowledge-based and data-driven AI techniques can be applied to detect leaks, correct errors, predict demands for predictive control with best energy efficiency, maximising productivity and helping digitise the NHS.