Novel Non-Invasive Detection of Thin Film Biofilm and Classification of Deposits Using Machine Learning

Abstract

Clean, safe, readily available water is vital for public health, irrespective of whether it is used for drinking, domestic use, food production, or recreational purposes. Globally, around two billion people use feces-contaminated water sources, which poses a high risk to the safety of drinking water due to the high probability of water contamination. Microbial-influenced corrosion is a significant problem in several industries, including but not limited to wastewater treatment, drinking water distribution systems, food industries, power plants, paper industries, and chemical manufacturing facilities. The presence of microorganisms causes around 70% of the corrosion in gas transmission pipelines, and corrosion accounts for the loss of around 4% of the gross national product. The United States is estimated to spend around $300 billion yearly on corrosion costs. A significant amount of time is spent finding and fixing the problem with a major overhaul or part replacement, saving about 30% of overhead costs. Due to its attachment, sanitization and cleaning methods are ineffective against biofilm in its mature stage. Overall, there is a need for a rapid assessment of pipes or other structures to assist in biofilm monitoring and cleaning procedures. The study presents and examines a fresh approach that combines non-invasive and non-destructive methods for detecting deposits in near real-time. The detection is accomplished by measuring changes in voltage and time-of-flight of ultrasound sensors and using a Random Forest machine learning algorithm to categorize the deposits into four types: no deposit, biofilm deposit, and corrosion deposit. This work builds a strong foundation for a future novel research approach for the detection of biofilm using evanescent waves or multiple internal reflections [1]Additionally, the technique is cost-effective, portable, and requires minimal power. Although random forest learning has been utilized for various classification problems, this study presents a novel application of the ML technique to classify deposits based on voltage and time of flight measurements. Unlike conventional methods like microscopic methods, combining the sensor arrangement with ML techniques allows users to make informed decisions on cleaning strategies, preventing massive biofilm buildup or other deposits in a closed wall piping system.