Detection of excessive atmospheric emissions at technological facilities

Determining emissions through smokestacks and industrial building envelopes

WORK PERFORMED

Software development

Designed for video analysis from four cameras to identify excessive emissions from smokestacks and non-organized sources at industrial facilities.

Signal transmission to HMI

Implemented signaling for weak and strong emissions at technological sites on the mimic diagram.

Mimic diagram development

Developed mimic diagrams for visual display of emission signals.

SYSTEM ALGORITHM

3 stages

Stage 1

Regions of interest and the feature vector for excessive emissions are identified in the image.

Stage 2

The vector is fed into a trained neural network that determines the presence and intensity of emissions at production facilities.

Stage 3

Based on neural network data, a signal is generated and sent to the mimic diagram, where it is converted into a visual display.

Mimic diagram overview

On the mimic diagram, the presence of signals from cameras and emissions at facilities are determined by color indication:

Green - no emissions
Yellow - weak emissions
Red - strong emissions
Gray - no signal from the camera

Hovering displays the time of recorded emissions.

Report histogram

Based on reports received from the system, histograms are generated characterizing the presence and degree of emissions on the battery.

The general view allows for a quick summary of emissions per day for each object.

RESULTS

Monitoring visible emissions over the territory

Using optical cameras and computer vision algorithms, the system continuously scans smokestacks, recording the moments when visible smoke appears. Monitoring is conducted 24/7, regardless of weather conditions (fog, rain) or time of day. This allows for the rapid detection of technological mode violations leading to excessive emissions and immediate signal transmission to responsible personnel.

Access to up-to-date information

Real-time emission data is sent to workstations via a web interface or corporate system. Information is updated at a set frequency (from 1 minute) and includes not only the fact of an emission but also an assessment of its intensity, duration, and link to a specific source. Authorized users can view history, generate charts, and export reports for regulatory authorities.

Detection accuracy up to 80%

The system's algorithms classify emissions into 3 types: weak (short-term), yellow (moderate), and strong (high pollution volume). The identification accuracy for each type reaches 80%, confirmed in real industrial production. Error is related to the variability of optical conditions (lighting, background), but the achieved accuracy is sufficient for early warning and taking measures without false positives. The system records the exact time each emission starts and ends, allowing for the calculation of the total volume of excessive emissions.

Back to projects Discuss a project

Client

Location

Coke Plant: Smokestacks

Year

2019