ML model and data processing pipeline for manufacturing
#Industry & Energy #Python #Azure #Apache #OpenCV #PyTorch #Energy
Project definition
Location
Client
Project type
Industry
Service list
ML Engineering / Computer Vision
Team size
S-Size (1–3 engineers)
Budget
$10,000 – $50,000
Task
The main task of this project was to develop an ML model capable of detecting and annotating smoke coming from factory chimneys in images. In addition to the model itself, a data pipeline needed to be implemented to process a large number of surveillance camera images in real-time for smoke detection. The model was trained using manually annotated data with images of smoke coming from chimneys in various weather conditions
Solution
The model and data pipeline were successfully implemented at various client factories in North America. The solution enabled real-time monitoring of factory conditions, alerting to unforeseen situations related to the emission of harmful substances into the air, as well as detecting emergency situations. With a model accuracy of 81%, the system reduced the number of incorrect decisions made by personnel and allowed the implementation of an external automated system for starting and stopping production based on the situation assessment
Impact
The implementation of the computer vision system delivered immediate improvements in environmental safety and operational efficiency. The solution, which processes feeds from 150+ cameras in real-time with 81% accuracy, reduced incorrect operator decisions by over 40% and enabled automated production shutdowns during critical emission events. This significantly lowered the risk of regulatory penalties and prevented costly operational accidents
💡 This is an AI-CORE project
💡 This is an AI-powered project
💡 Have a similar Industrial or Energy request?
💡 Have a similar Healthcare request?
💡 Have a similar EdTech request?
💡 Have a similar FinTech request?
💡 Have a similar request?
Data processing from 150 cameras
Developed a data pipeline capable of processing a large number of images from 150 cameras in real-time
The accuracy of the ML model recognition is 81%
Development Process
We built a computer vision model using PyTorch, OpenCV, and gradient boosting methods, trained on manually annotated images of chimneys in diverse weather conditions. A batch and real-time inference pipeline was created with Azure services, including pre- and post-processing for large-scale image flows. Apache tools were employed for handling high-throughput data streams, while Azure pipelines ensured scalable deployment and monitoring. The system was designed for integration into factory control systems, providing reliable smoke detection and actionable insights to automate safety responses
Technologies
ML model and data processing pipeline for manufacturing
Image analysis using ML to detect harmful emissions in manufacturing
Book a call with our Head of Sales
