Tuesday, 11 October 2022
Early detection and intervention of gas leaks can reduce industrial risks. Numerous early detection methods have been developed and applied in the industrial field, but they require professional analysis technology, long analysis time, and relatively expensive equipment. Strategies to reduce industrial gas hazards constantly require the development of new, efficient and reliable devices.
In the past few decades, various types of gas sensors have been developed based on semiconductors, optical fibers, acoustic waves and other technologies. A significant amount of work has been done in recent years to make sensors smaller, more sensitive, more visually responsive, more manageable, more personal, and more wearable. However, these functions require expensive equipment, complex skills, and large amounts of power. In this regard, colorimetric sensors are of great interest due to their simplicity, ease of manufacture, low cost, visual response, and ease of interpretation without power consumption. However, current colorimetric sensors have some limitations to overcome, such as sensitivity, high cost, early detection and accuracy for use in industrial safety systems. Conventional colorimetric sensor studies use algorithms to compare discoloration with the naked eye, estimate gas concentrations with simple statistical methods, or estimate simple discoloration information using machine learning algorithms to estimate time-series data. Algorithms are trained on measured images or time series data extracted from images. Signal processing focuses on denoising done by scaling time series data. These studies using machine learning and deep learning-based analytics for colorimetric sensor studies did not give sufficient consideration to data coding. Machine learning of time series data is performed using a time series decomposition method that separates time dependencies. I am trying to analyze time-series data through image encoding.
In this study, a time series data analysis technique using CNN (Convolutional Neural Network) including multidimensional image encoding was applied to improve the gas concentration estimation accuracy. The estimation accuracy obtained in this study was higher than the estimation accuracy obtained by applying existing statistical methods such as Auto-Regressive Integrated Moving Average (ARIMA) and Exponential Smoothing (ETS) and machine learning techniques such as Multi-layer Perceptron (MLP).
We focused on estimating gas concentrations using a time-series-based colorimetric analysis method developed based on colorimetric fabric detection data. Time series data obtained through gas experiments were encoded into images to classify the color value change trend of gas concentration. Image encoding improves the efficiency of specifying gas-exposed sections and estimating gas concentrations within the entire section. Compared to the concentration estimation accuracy of the existing data classification method, MLP was improved by more than 2.5%. In addition, the accuracy was improved by applying a CNN that learned multidimensional processing of sensor data through image encoding. As a result, it was confirmed that the estimation accuracy was improved by 6.1% compared to the basic estimation technique. Although this method has limited data form and application fields, it is effective for subjects such as ours. The proposed technology can be applied to the next-generation smart industrial safety system.
In the past few decades, various types of gas sensors have been developed based on semiconductors, optical fibers, acoustic waves and other technologies. A significant amount of work has been done in recent years to make sensors smaller, more sensitive, more visually responsive, more manageable, more personal, and more wearable. However, these functions require expensive equipment, complex skills, and large amounts of power. In this regard, colorimetric sensors are of great interest due to their simplicity, ease of manufacture, low cost, visual response, and ease of interpretation without power consumption. However, current colorimetric sensors have some limitations to overcome, such as sensitivity, high cost, early detection and accuracy for use in industrial safety systems. Conventional colorimetric sensor studies use algorithms to compare discoloration with the naked eye, estimate gas concentrations with simple statistical methods, or estimate simple discoloration information using machine learning algorithms to estimate time-series data. Algorithms are trained on measured images or time series data extracted from images. Signal processing focuses on denoising done by scaling time series data. These studies using machine learning and deep learning-based analytics for colorimetric sensor studies did not give sufficient consideration to data coding. Machine learning of time series data is performed using a time series decomposition method that separates time dependencies. I am trying to analyze time-series data through image encoding.
In this study, a time series data analysis technique using CNN (Convolutional Neural Network) including multidimensional image encoding was applied to improve the gas concentration estimation accuracy. The estimation accuracy obtained in this study was higher than the estimation accuracy obtained by applying existing statistical methods such as Auto-Regressive Integrated Moving Average (ARIMA) and Exponential Smoothing (ETS) and machine learning techniques such as Multi-layer Perceptron (MLP).
We focused on estimating gas concentrations using a time-series-based colorimetric analysis method developed based on colorimetric fabric detection data. Time series data obtained through gas experiments were encoded into images to classify the color value change trend of gas concentration. Image encoding improves the efficiency of specifying gas-exposed sections and estimating gas concentrations within the entire section. Compared to the concentration estimation accuracy of the existing data classification method, MLP was improved by more than 2.5%. In addition, the accuracy was improved by applying a CNN that learned multidimensional processing of sensor data through image encoding. As a result, it was confirmed that the estimation accuracy was improved by 6.1% compared to the basic estimation technique. Although this method has limited data form and application fields, it is effective for subjects such as ours. The proposed technology can be applied to the next-generation smart industrial safety system.