Automatic Classification of Construction Site Noises - ACOEM Partnership
Noise pollution caused by construction sites in urban areas can quickly become disturbing and be a source of complaints from local residents. It is then a major issue for HSE managers and for project managers.
Equipped with a fleet of more than 80 sound level meters and its team of acousticians, Cementys now offers on-site acoustic tracking to monitor sound intensity thresholds and to explain why they are exceeded by listening to the recordings. In order to improve current acoustic monitoring and provide an innovative solution, Cementys and ACOEM are developing an automatic sound classification solution based on AI techniques. This solution will allow for the construction sites to be warned by SMS and email, as soon as certain thresholds are exceeded.
The project is in two phases:
- The first phase of solution research and construction of the worksite dataset to validate the classification,
- The second phase in partnership with ACOEM, one of the world leaders in acoustics, to develop the automatic classifier and then to embed the intelligence directly into the sound level meters.
With this partnership, Cementys implements its partnership strategy to innovate by creating win-win ecosystems with its suppliers.
Phase 1 – Dataset Construction and the First Test:
In order to carry out the first tests of automatic classification of exceedances, environmental noise data sets were used. Among these environmental noises, we can find construction site machinery that can be the cause of exceedances (drill, truck engine, alarm siren).
At the same, the Cementys Acoustic compile a set of site noise data from sound level meters in real conditions. For the labeling of these data, the teams relied on the date of the exceedances in combination with site reports to determine the exact source. The worksite labels were chosen based on experience in the field and the “Silence Chantier” white paper provided by the Société du Grand Paris (SGP) to limit noise pollution on the worksite.
Currently the dataset contains 26 worksite labels and more than 1032 excerpts of exceedances. The automatic classification solutions tested by spectrogram extraction allow to obtain satisfactory results (more than 75% success rate) on the surrounding sound datasets.
Phase 2 – Partnership with ACOEM:
In order to improve the solution’s first test phase and to embark it directly into the sound level meters, to perform a real-time classification, Cementys and ACOEM (designer of sound level meters and French leader in acoustic analysis) have established a partnership.
The final solution will be integrated directly into the sound level meters to perform the classification at the time of detection of exceedances by the sound level meter. Thus, this will be an added value for Cementys and its acoustic monitoring offer, as well as for ACOEM, which will be able to offer a new functionality on its sound level meters.