In addition to lengthy and expensive operations research studies, factory production can achieve its optimal goals through comprehensive analysis and monitoring of the entire system. Furthermore, production workflows are efficient when machines are working at peak efficiency without error intervals or exceptions.
Traditional machine life assessment methods typically focus on diagnostics and lack the ability to predict failures or threats. Therefore, the integration of artificial intelligence provides a solution by introducing the concept of machine-level predictive analytics.
Often, systems fail due to deterioration of components or any external force that interferes with the overall process. Therefore, it can be interpreted that there must be some abnormal behavior of the system to crash. In most cases, these anomalies occur gradually rather than suddenly, making it possible to detect and predict anomalies before they have a significant impact on the system.
Advantech Edge Nodes for Sensor Data Collection
Data collection plays an important role in performing exploratory machine-level analysis and then fitting the dataset to an AI model for anomaly detection. A range of sensors are required to measure the physical characteristics of the machine to assess condition and performance.
Reality AI and Advantech have teamed up to develop an edge node for evaluating machines used to detect anomalies and predict the expected lifespan of operating components. It is named RealityCheck AD edge node based on Advantech EPC-S201 fanless embedded PC. Intel's Celeron N3350 dual-core SoC forms the heart of the embedded PC, with 8GB of RAM and a 64G SSD, showing a wide range of processing and storage capacity.
It comes with a wide range of sensor options for collecting data from the machine:
Accelerometers from different manufacturers
Current and temperature sensors for performance and thermal evaluation
Contact Microphone for Mechanical Audio Sensing
Edge nodes are available with wall-mount and rail-mount options for flexible deployment, matching most deployment possibilities in the industry. In addition, the device supports a wide range of connectivity, including support for Wi-Fi, Ethernet and cellular communications.
Anomaly detection using Reality AI software tools
Sensor data needs to be cleaned and preprocessed before any AI algorithms can be applied. Reality AI software efficiently extracts features from data after thorough preprocessing and cleaning. It uses an algorithmic search with an extensive feature space to derive custom transformations based on time and frequency. The software provides the best results in the case of anomaly detection, signal classification and life expectancy prediction.
Reality AI also has the ability to adjust AI model weights based on the use case. It can also create machine learning models and data visualizations to improve the accuracy and understanding of applications. Additionally, it provides in-depth hardware analysis that aids solution design.
RealityCheck AD comes with a baseline normal policy for optimizing anomaly detection. The method evaluates reference regions in feature space and compares observations to baseline normal regions (actually reference regions). Reality AI software predicts the correct feature space for baseline normals by classifying tightly compressed regions. All known anomalies are far away from the edge of the baseline, so distant points become anomalies, completing the detection process.
The anomaly detection process begins by creating a baseline region in the initial detection model. The real-time data collection is continuously updated in the dataset, while the deployed detection models predict anomalies in the system. Investigate and verify this forecast to determine if the forecast is completely accurate. In the case of true predictions, specific data points are marked as anomalies and updated in the dataset. Likewise, in the case of mispredictions, the data is labeled and updated accordingly.
Although the AI model is accurate, there may be some anomalies that may go undetected due to noise or overfitting. But most of them are detectable in equipment monitoring and end-of-line testing. Since equipment monitoring is the foundation of machine-level analysis, insight into failures can provide significant cost savings for overall production. In addition, end-of-line testing ensures quality and performance and is also important from a safety perspective. RealityCheck AD edge nodes enhance production machines to meet optimal goals.
Reviewing Editor: Guo Ting