From data to action: Transforming pressure testing in manufacturing with machine learning for enhancing energy efficiency

Abstract

In today's competitive market, manufacturing in energy-efficient facilities is crucial for a product's success. High energy costs can severely impact product competitiveness, making it challenging for inefficiently produced goods. By carefully evaluating data collected throughout the production process, it's possible to reduce energy consumption without sacrificing product quality, thus achieving cost savings, and ensuring the economic and environmental sustainability of industrial operations. Using machine learning methodologies, which are excellent at analyzing large data sets, can significantly enhance process efficiency. Optimizing energy use through these algorithms not only cuts costs but also reduces environmental impact. Machine learning usage is beneficial at minimizing production errors and utilizing resources more effectively, helping industrial facilities gain a competitiveness while implementing eco-friendly solutions. This approach supports the transition to greener, more sustainable industrial activities, for smarter, more efficient, and environmentally operations. This study investigates the use of machine learning techniques to enhance the efficiency of pressure drop tests, which are conducted at different stages of production processes to measure the accuracy of the product or process. Traditional pressure testing methods pose significant challenges for industrial enterprises in terms of cost and environmental impact due to their high energy consumption and lengthy durations. These tests are also used for leakage detection to ensure product safety and structural integrity. This thesis examines the applicability and effectiveness of advanced machine learning algorithms to optimize pressure testing processes and enhance the energy efficiency and decrease the cycle times. Over the course of more than a year, a large dataset containing 1.7 million test records was utilized from a production plant testbenches. Data firstly classified and merged based on standardized data analysis systematics. Afterwards, the data was used to deeply analyze the dynamics of the test processes, its internal correlations regard to paramaters and identify potential areas for improvement especially focused on duration decrease of tests for efficiency reasons. The study developed a model using various machine learning algorithms such as Random Forest Regressor, XGBoost, and Linear Regression to predict the optimum duration of pressure drop tests. Each algorithm has been optimized using cross-validation techniques. The dataset was randomly divided into two groups, 75% for training and 25% for testing. Here, machine learning algorithms were executed, and the performance of the models was examined and compared using statistical indicators based on the root of the sum of the squares of the differences between the actual values and the values predicted by the models, such as Root Mean Square Error (RMSE) and R-squared (R2). Subsequently, the split of the test and training datasets were swapped in the model for cross-validation purposes, and the performance of the model was tested for different random splits. Although cross-validation caused the model to get results very close, to ensure once more that this was not random and to complete the validation, the dataset was divided into three groups (train, test, validation).