In many industries, analog gauges are monitored manually, thus posing problems, especially in large facilities where gauges are often placed in hard-to-access or dangerous locations. This work proposes a solution based on a microcontroller (ESP32-CAM) and a camera (OV2640 with a 65° FOV lent) to capture a gauge image and send it to a local computer where it is processed, and the results are presented in a dashboard accessible through the web. This was achieved by first applying a Convolutional Neural Network (CNN) to detect the gauge with the CenterNet HourGlass104 model. After locating the dial, it is segmented using the circle Hough transform, followed by a polar transformation to determine the pointer angle using the pixel projection. In the end, the indicating value is determined using the angle method. The dataset used was composed of 204 gauge images split into train and test sets using a 70:30 ratio. Due to the small size of the dataset, a diverse set of data augmentations were applied to obtain high accuracy and a well-generalized gauge detection model. Additionally, the experimental results demonstrated adequate robustness and accuracy for industrial environments achieving an average relative error of 0.95%.
The emergence of Industry 4.0 and 5.0 technologies has enabled the digital transformation of various processes and the integration of sensors with the internet. Despite these strides, many industrial sectors still rely on visual inspection of physical processes, especially those employing analog gauges. This method of monitoring introduces the risk of human errors and inefficiencies. Automating these processes has the potential, not only to boost productivity for companies, but also potentially reduce risks for workers. Therefore, this paper proposes an end-to-end solution to digitize analog gauges and monitor them using computer vision through integrating them into an IoT architecture, to tackle these problems. Our prototype device has been designed to capture images of gauges and transmit them to a remote server, where computer vision algorithms analyze the images and obtain gauge readings. These algorithms achieved adequate robustness and accuracy for industrial environments, with an average relative error of 0.95%. In addition, the gauge data were seamlessly integrated into an IoT platform leveraging computer vision and cloud computing technologies. This integration empowers users to create custom dashboards for real-time gauge monitoring, while also enabling them to set thresholds, alarms, and warnings, as needed. The proposed solution was tested and validated in a real-world industrial scenario, demonstrating the solution’s potential to be implemented in a large-scale setting to serve workers, reduce costs, and increase productivity.
Visual inspection based systems are important tools to ensure the quality of manufactured parts in industry. This work presents an automatic visual inspection approach for defect detection in turbo vanes in the investment casting industry. The proposed method uses RANSAC for robust line and circle detection to extract relevant information to discriminate between a good part and a defected one. Then, using this data a feature vector is created serving as input to a SVM classifier that after the training phase is able to discriminate and classify between a good sample or not. To test the proposed approach a private database was created containing 650 turbo vanes (which gives 2600 different samples to train and test). On this database the proposed method achieved an average accuracy of 99.96%, an average false negative rate of 0.00% and an average false positive rate of 0.05%, using a 5-fold cross validation protocol, which demonstrates the success of the proposed method. Moreover, the proposed image processing pipeline was deployed into Raspberry Pi 4 Model B part of a visual inspection machine, and is working daily at ZCP – Zollern and Comandita Portugal, which proves the method's robustness.