Barriault, MatthewMontazeri, Mahyar MohagheghO'Brien, AllenNajjaran, HomayounHoorfar, Mina2018-11-082018-11-08May-18978-1-77355-023-7http://hdl.handle.net/10315/35334http://dx.doi.org/10.25071/10315/35334Paper presented at 2018 Canadian Society of Mechanical Engineers International Congress, 27-30 May 2018.In order to detect natural gas pipeline leaks, ethane in the natural gas must be discriminated from background methane emissions. Our gas detection apparatus is well-suited for this application due to its flexibility and low cost. We present a comparison of machine learning models for quantitative estimation of concentrations of both methane and ethane in a target gas sample, using a response over time from a single sensor in our apparatus. We also demonstrate that the use of synthetic data is very effective for training a model to discriminate between methane and ethane.enThe copyright for the paper content remains with the authorMicrotechnology and NanotechnologyMachine learningGas detectionMicrofluidicDiffusion simulationPipeline leak detectionArticle