Search

International audience

International audience; Indoor air quality (IAQ) is a growing and multidisciplinary research field domain involving professionals with various expertise, from Materials Sciences, Instrumentation, Electronics, Physics, Chemistry and many more. One of the current hot topics of this field is to equip private accommodation with air quality monitors. For that, it is crucial to develop and choose the more sui table sensors capable of detecting and discriminate different types of gases and volatile organic compounds. The number of needed sensors is related to which substances are going to be monitored. However, this choice is not obvious as the intrinsic information of sensors is not always well understood. The main objective of this work is to understand and provide a framework on how this choice can be made. Several different types of signals are generated when using multisensory systems, but not all the signals are useful for detecting the aimed substances. By analyzing these signals, it is possible to select the sensors presenting the best match with a given substance detection task. To analyze the collected data, a technique of intrinsic dimension estimation was used. This technique allows the estimation of how many dimensions (independent variables) are necessary to synthesize equivalent signals. In the case of multisensory systems, the dimensions can be interpreted as the number of sensors present in it. However, the intrinsic dimension estimation of a dataset collected via multisensory systems may reveal a smaller effective dimension, thus possibly allowing the estimation of the minimum number of sensors needed for a given task. The dataset analyzed in this work is composed by the responses of 40 RUBIX-PODs with 19 sensors each, during 6 experiments where specific gas injections are done to represent different indoor activities (cleaning, cooking, etc.). The intrinsic dimension (ID) of this dataset is estimated. The data was arranged accordingly to an injection discrimination task, i.e., detecting what experiment (here called injection) generated each data point. Preliminary results showed an ID of 3 for the given dataset and indeed when only three specific sensors were used, all the injections could visibly be separated. Although this result shows that ID estimation is a good indication of the number of sensors needed, this experiment is very simple compared to what the multisensory systems face during day-to-day use, more experiments using real world data collected by those systems should provide a better understanding on what they are detecting and what sensors are really needed.

International audience