MEDICAL GAS SYSTEMS
Data Trending and Prediction Critical Parameters
Pressure sensor
Flow sensor
Signal collector/ controller (PCB)
4G/5G RS485
Current sensor
Plant Alarm panel
Temperature sensor
Machines
Real-time AI System The real-time AI system has been designed following Industry 4.0 principles for MGSs remote monitoring and intelligent plant control (as shown in Figure 1). In the real-time AI system, printed circuit boards (PCBs) are designed as machine controllers, signal collectors, and transmitters, and the IoT communication networks are built for real-time monitoring, control, and AI optimisation for highly efficient operation of MGSs. With the IoT networks built (RS485 and wireless communication protocol developed, 4G/5G mobile networks), real-time fault prediction and detection based on multi-sensor measurements can be performed by deep learning, providing an eagle eye view of every event while or before it is happening. With the data collected over IoT networks and streamed to a cloud server that can run more detailed data processing for, as an example, predictive control by the AI algorithm, the hybrid edge and cloud approach developed can now provide edge autonomy as well as support remote monitoring and intervention when required. It also enables real-time big data analytics to improve decision-making and performance with minimised energy usage and waste, creating an opportunity to apply deep learning for advanced control, management, and maintenance by collecting data from all sources on the IoT networks. Through real-time processing of measurement data provided by dedicated sensors installed, autonomous decision-making is enabled based on the online diagnosis of the system with right conditions, leading to increased reliability toward zero defects. By collecting, analysing, and making use of data, predictive maintenance based
IFHE DIGEST 2023 Medical air plant Figure 1. Real-time AI system designed following Industry 4.0 principles.
on real-time AI can also be planned and scheduled.
As a machine learning algorithm, deep
neural network (DNN) has shown powerful performance in a variety of applications by processing large datasets. However, slow learning, convergence to a local minimum, over-parameterization, high computational complexity, and hardware resource requirements remain as technical challenges in such a network, meaning that the existence of overfitting in association with over-parameterisation and presence of noise can worsen the performance of a network and make real- time applications difficult to implement. In addition, initialization leading to exploding or vanishing gradients, and the bias variance trade-off property of a DNN model with high complexity, where variance can be attributed to overfitting inconsistencies in the labels due to noise in the training set, both require further investigation. To address these challenges and issues for better network performance and faster computation of the real-time AI system, Mexican hat (Gaussian) wavelet with tight frame and sparse representation as activation functions thus has been derived to construct novel DWNNs. It is the negative normalized second derivative of a Gaussian function and has admissibility condition and a symmetric structure. And its derivative is also a Gaussian function. Because of the properties, such as optimal time-frequency localisation, admissibility, and energy preservation, Gaussian wavelet functions are advantageous in signal detection, feature selection, and data decomposition. In the context of DWNNs, input data is decomposed into the time-frequency domain by the wavelet activation functions in the neurons and connected by weights, which
can be interpreted as wavelet coefficients corresponding to the input or output signals that can be approximated by the fitted signal and thus the initial values of the weights can be obtained from wavelet coefficients, which are generated to sparsely represent the relationship of input and output in such a way that training process can converge quickly and training error can be globally minimised. Due to the sparsity of the wavelet activation function, the designed network structure can also be simplified and optimized to reduce the risks of network overfitting and computational complexity. It also allows the network bias and output variance to be minimised at the same time. Compared with commonly used deep neural networks, using nonlinear Gaussian (Mexican hat) wavelet as activation functions in the DWNNs model can improve nonlinear fitting and convergence speed with adaptive learning rate, so that the power and capabilities of machine learning can be raised and accurate predictions can be made by the trained DWNNs with faster learning speed. In addition to enabling predictive control and faults prediction, the real-time AI system can also be trained to learn and diagnose what cause the faults and what actions can be taken to prevent the failure or reduce system damage. The development of the real-time AI
system with the trained DWNNs to deploy Industry 4.0 thus allows the implementation of predictive control and maintenance programs for improving energy efficiency and achieving optimal performance of MGSs.
Experiments and Discussions Using real-time AI, where the sensing capabilities and the computational power are provided by the designed controller,
37 Manifold User interface (Tablet/PC, mobile phone) IoT gateway All Normal
Cloud
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68 |
Page 69 |
Page 70 |
Page 71 |
Page 72 |
Page 73 |
Page 74 |
Page 75 |
Page 76 |
Page 77 |
Page 78 |
Page 79 |
Page 80 |
Page 81 |
Page 82 |
Page 83 |
Page 84 |
Page 85 |
Page 86 |
Page 87 |
Page 88 |
Page 89 |
Page 90 |
Page 91 |
Page 92 |
Page 93 |
Page 94 |
Page 95 |
Page 96 |
Page 97
