Client Background
The client is a company operating within the energy manufacturing industry. Due to a Non-Disclosure Agreement (NDA), further details regarding the specific company cannot be disclosed.
Challenges
The client had two primary goals:
- Improve data quality: Data was being manually recorded in notebooks and then transcribed into the databases, leading to potential errors and inaccuracies.
- Enhance worker safety: The manual data collection process involved navigating industrial facilities, often requiring climbing ladders and accessing difficult-to-reach locations, posing a risk of workplace injuries.
Our Solution
Automating Measurement Device Readings
To improve data quality and worker safety, a mobile app was developed for ruggedized Android devices. These devices, while durable, have limited specs (older Android versions, lower camera quality, and limited memory). The app was designed to be user-friendly yet sophisticated enough to recognize various analog measurement devices.
Recognizing Analog Measurement Devices
The key challenge was accurately recognizing analog devices (e.g., pressure gauges, thermometers) despite camera limitations. Given the devices’ low processing power, traditional computer vision techniques were used instead of deep learning.
The recognition process involved:
- Key Element Detection: Identifying circles, scales, tick marks, min/max values, and the needle.
- Trigonometric Calculation: Determining the reading based on the needle’s position.
- Addressing Challenges: Handling negative scales, reflections, poor lighting, and obstructions using domain expertise.
QR Codes for Device Identification
Each device was assigned a unique QR code to ensure accurate identification. Scanning the code before taking a photo provided location, identifier, and calibration data, crucial for managing devices across different facility locations.
Integration and Accuracy
Data was transmitted via the tablet’s SIM card and stored in a database. The goal was at least 80% accuracy in automated readings, with manual entry available for uncertain cases. Testing confirmed this accuracy was achieved.
At a Glance
The client, operating in the energy manufacturing industry, struggled with data inaccuracies and safety risks due to manual data collection. To solve this, we created a mobile app for ruggedized Android devices that automates gauge readings using computer vision. By recognizing key elements on analog devices and calculating readings based on needle positions, the app improved data accuracy. QR codes enabled precise device identification, ensuring efficient data.
