Created on 26th September 2024
•
Our AI-powered quality control platform automates and enhances manufacturing processes by integrating data from IoT sensors, cameras, SCADA systems, PLCs, and SAP, supporting protocols like OPC-UA, Modbus, and MQTT. In scope for the solution are features such as real-time anomaly detection, predictive maintenance, and advanced pattern recognition to minimize defects and ensure consistent product quality. Our GenAI-powered Machine Knowledge System (MKS) facilitates root cause analysis by allowing users to upload OEM manuals and other process documents, linking alerts to potential causes for rapid resolution. Additionally, the platform offers in-house developed computer vision models for live defect detection, enabling manufacturers to monitor quality in real time.
Out of scope are hardware installations, extensive custom software development beyond the core features, and manual inspection processes, which are effectively automated by our solution. The focus remains on software integration and analytics rather than physical equipment upgrades or replacements.
Future opportunities for this solution include expanding its capabilities into other manufacturing sectors, enhancing the platform with advanced machine learning algorithms for better predictive insights, and developing partnerships for comprehensive supply chain integration. Moreover, we envision integrating additional data sources and protocols, refining the MKS for broader applications in compliance and sustainability, and scaling the solution to support small and medium enterprises (SMEs) in adopting digital transformation. By continuously evolving our technology, we aim to help manufacturers achieve operational excellence and contribute to India’s vision of becoming a global manufacturing hub.
One significant hurdle we encountered was integrating existing CCTV cameras into our AI-powered quality control platform, particularly with analog camera feeds. Initially, the challenge lay in effectively routing the analog video streams through our RTMP server, which was essential for applying our computer vision models within the machine learning pipelines. The standard method of processing these feeds resulted in latency issues and inconsistent data quality.
To address this, we developed a specialized device that captures the analog feed and converts it into a digital RTMP stream. This conversion process involved using an embedded system with encoding capabilities that could handle real-time video processing while ensuring minimal delay. Once the analog feeds were digitized, we utilized Apache Airflow to orchestrate our ML pipelines, allowing for efficient scheduling and monitoring of the custom training workflows for our defect detection models. We implemented a continuous integration and deployment (CI/CD) approach to streamline updates and maintain the flow of data through the pipeline. This setup allowed us to regularly retrain our models with new data, enhancing their accuracy over time.
Tracks Applied (1)