Machine learning in botda fibre sensor for distributed temperature measurement

• Main Author: Nur Dalilla binti Nordin
• Format: text::Thesis
• Language: English
• Published: 2023
• Subjects: BOTDA FIBRE SENSOR; DISTRIBUTED TEMPERATURE

Power transmission cable is an essential asset in the energy industry, and thus, it is crucial to make sure that the cables are working optimally at all times. One of the ways to monitor the performance of the transmission cable is by monitoring the temperature. It is known that the conductor temperature increases proportionally to the current loading injected into the cable. The protecting sheath of the power cable, however, has its limited operating temperature before it breaks down. Therefore, a real-time temperature sensor would be a preferable option to be introduced as a method to monitor the operation condition of any power cable in order to improve performance. To make the sensor more reliable, temperature data must be collected over the length of the cable, or distributed data rather than point data. One of the methods that were proven to be successfully deployed in many industrial sections is fibre-based distributed sensing. Fibre optic cable laid alongside power cable may be used as a temperature sensor as temperature affect light scattering in the fibre and can be measured. In this thesis, the employment of one type of the scattering-based methods which is known as the stimulated Brillouin scattering (SBS), by using the Brillouin Optical Time Domain Analyser (BOTDA) technique is used to acquire the temperature profile along the whole length of the fibre. BOTDA utilizes two counter-propagating signals, pulsed pump- and continuous-wave probe lights to generate SBS in fibre. When the frequency difference between the two lights is tuned to the local Brillouin frequency shift (BFS), maximum amplification occurs. The shift in BFS, however, will occur due to external perturbations such as changes in temperature acting directly on the fibre. Conventionally, the Lorentzian curve fitting (LCF) method is deployed in the BOTDA technique to construct the Brillouin gain spectrum (BGS) and consequently extract the BFS. However, the retrieved spectra are often noisy and thus affect the whole process of determining the BFS. The accuracy of BFS calculated based on LCF is highly dependable on the initial parameter setting of the curve fitting process. Besides, the total time taken using the method is relatively long, especially when measuring long fibre. An alternative method is proposed, utilizing machine learning algorithms. Therefore, this thesis explores the comparative analysis for BOTDA data processing using the six most suited machine learning algorithms. By utilizing machine learning, the need to find BFS can be eliminated and thus further reducing the total processing time. It was found that machine learning algorithms have significantly reduced the signal processing time by 3.5 to 655 times faster than the conventional LCF method. Additionally, temperature prediction accuracy and temperature measurement precision made by some of the chosen algorithms were comparable, and some were even better than LCF. The results obtained in these experiments would provide some overview in deploying machine learning algorithm for characterizing the Brillouin-based fibre sensor signals.