IoT Platform Enhanced With Neural Network for Air Pollutant Monitoring

This work presents the design and setup of an IoT platform at level four of the technology readiness level (TRL-4) to detect, classify, and quantify pollutant gases. This study combines concepts such as wireless sensor networks (WSNs), arrays of sensors, and multivariate data analysis to interface different nanostructured chemiresistor gas sensors. The IoT platform consists of several gas sensor nodes (GSNs) with Wi-Fi capability to send data from a sensor array to a server and its user interface (UI). Each GSN interfaces one sensor array (up to four chemiresistor gas sensors and one temperature and humidity sensor). The server channels the data from the GSNs to the UI. The platform was set up following a two-stage methodology. First (training stage), sensor data were received, stored, and used to train different multilayer perceptrons (MLPs) artificial neural networks (ANNs). Second (recognition stage), models were implemented in the UI to classify and quantify the presence of pollutants. The platform was tested in laboratory conditions under exposure to nitrogen dioxide and ammonia at a different %RH. As a result, the platform improves the classification and quantification times compared with the single-sensor approach. In addition, the system was evaluated using a gas mixture of both gases, showing a classification accuracy exceeding 99%. Likewise, the training and recognition stages can be repeated to add new chemiresistor gas sensors in the node, add new nodes to the platform, and deploy the nodes in different scenarios.