MEDICAL GAS SYSTEMS
DWNN network for predicting faults. The training was performed offline to obtain the optimal network parameters. Tests can be performed online to predict the timing of system status in four stages: normal, first warning, second warning, and fault detected. An example is shown in Figure 4,
where four stages of system status were identified, and features extracted were used for training the DWNN to predict and make decisions. It is noted that before faults were detected, there are two patterns that signify large changes of system status. At the same time, values of oxygen purity and oxygen line pressure keep decreasing. These can be classified as first warning (start at data point 51948) and second warning (start at data point 52446). At data point 53284, the fault is
detected and alarm triggered following features indication and the oxygen purity at a level lower than a pre-set threshold. Then faulty condition continues and, at data point 58908, it reaches the worst case when the values of oxygen purity and oxygen line pressure are both at their minimums. After that, values of both oxygen purity and oxygen line pressure are increasing, but still the system status is not getting back to normal. Between 62400 and 63097, compressors stop, and this is marked as downtime. Then compressors restart and eventually, at data point 63856, flow rate has dropped, and then the system is recovered and back to normal. The presence of faults in the MGS is
verified in an unsupervised manner by extracting signal features. The system will also diagnose each fault and perform cause analysis based on importance ranking explained by the trained DWNN model, where a fault can be related to a potential or unique type of component fault, and even to more than one fault, and then a knowledge-based system will be established to perform the explainability, from which the cause is analysed and a solution is sought. For example, the above detected faults can be attributed to very high demand of oxygen during the pandemic; when the flow rate increased to a very high level, the system could not generate enough output due to the limited capacity and led to a system failure. Before the fault was detected, there were many features/signs that indicated the trending and thus warnings were given and a decision on actions to be taken was made by the knowledge-based machine learning system.
Experimental results have
demonstrated that by the trained DWNN with multi-sensor measurements over the IoT networks, the proposed real-time machine learning system can
74
Figure 4. Details of multi-sensor measurement signals in time-frequency domain.
give valuable insights, track everything that is relevant to operations, and perform qualitative and quantitative data analysis to monitor machine conditions, detect faults, and predict and optimise system performance. Data analytics by deep learning can also help establish a knowledge-based system from which to correct errors and perform predictive maintenance procedures to prevent the failure without unnecessary interruptions.
Conclusions Due to the difficulty in interpreting the multi-sensor data correctly and extracting the desired information, and the strong nonlinearity and its nearly instantaneous response to disturbances, it is still very challenging for systems equipped with many sensors over IoT networks to achieve accurate fault prediction and optimised performance in complex multiple-input multiple-output (MIMO) systems.
In this paper, a new AI model is
proposed to improve the way that real- time fault prediction can be achieved by transforming the raw multi-sensor data into the time-frequency domain and deploying fast lifting wavelet transform with computational efficiency to obtain a big feature vector containing all the relevant features from all of the sensors and giving more informative signatures of system status. To predict and detect system
conditions accurately, a knowledge-based machine learning system is developed to track everything that is relevant to operations, where the sensing capabilities and the computational power are provided by the designed controller, transmitter and cloud server to enable confident decision making for fault prediction. By building IoT networks to remotely
collect data and developing deep wavelet neural networks (DWNN) with Gaussian
(Mexican hat) wavelet derived as activation functions to improve nonlinear fitting and convergence speed, the real- time data analytics in time-frequency domain can help establish a knowledge- based system from which to monitor conditions and optimise performance and predictive maintenance. Experimental results demonstrate that
features extracted can reveal the presence of a fault, and its type and cause can be explained by the trained DWNN over the IoT networks, so that by deep learning- based remote monitoring, the MGS systems can predict faults, correct errors, and maximise efficiency.
References 1 Luo G, Luo Y, Gan H. Predictive control with energy efficiency enabled by real-time machine learning, Proceedings of 1st IFSA Winter Conference on Automation, Robotics & Communications for Industry 4.0, 3-5 February 2021, Chamonix-Mont-Blanc, France, 007:34-38.
2 Qiu T, Chen N, Li K, Atiquzzaman M, Zhao W. How can heterogeneous internet of things build our future: a survey. IEEE Commun Surv Tutor 2018; 20: 2011-27.
3 Patan K. Neural network-based model predictive control: fault tolerance and stability. IEEE Transactions on Control Systems Technology 2015; 23 (3): 1147-55, .
4 Krishnamurthi R, Kumar A, Gopinathan D, Nayyar A, Qureshi B. An overview of IoT sensor data processing, fusion, and analysis techniques. Sensors 2020; 20: 6076.
5 Draeger A, Engell S, Ranke H. Model predictive control using neural networks. IEEE Control Systems 1995; 15 (5): 61-6.
6 Barmada S, Musolino A, Raugi M, Rizzo R, Tucci M. A wavelet based method for the analysis of impulsive noise due to switch commutations in power line communication (PLC) systems. IEEE Transactions on Smart Grid 2011; 2 (1): 92–101.
7 Sweldens W. The lifting scheme: a construction of second generation wavelets 1998; SIAM J Math Anal 29 (2): 511-46.
IFHE DIGEST 2022
IFHE
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 |
Page 98 |
Page 99 |
Page 100 |
Page 101 |
Page 102 |
Page 103 |
Page 104 |
Page 105 |
Page 106 |
Page 107 |
Page 108 |
Page 109 |
Page 110 |
Page 111 |
Page 112 |
Page 113 |
Page 114 |
Page 115 |
Page 116