CRONOS-Temporal Pattern Recognition

CRONOS is a system designed to find unusual patterns in data that changes over time. It helps with tasks like predicting when machines might break down, checking on the health of equipment, and finding problems. A big challenge in these areas is not having enough examples of failures to teach a computer system. CRONOS aims to find these anomalies right away, even without a history of past failures. This helps solve the "cold-start problem," which is a common risk for new products.

Normally, setting up a system to find problems involves collecting a lot of data, marking specific issues, training a computer model, and then using it. CRONOS skips the long steps of collecting and labeling data. It works on a "zero-shot" idea, meaning it doesn't need a library of past failures or special training for each new situation to start detecting problems. The system is built to be predictable, so the same input will always give the same output. This is important for results that can be checked and repeated.

While CRONOS might not always be better than specialized computer models that have been carefully trained on specific data, it offers other advantages. It helps you get useful results quickly when you don't have labeled data. It can also run on smaller computers, like those found in some devices, without needing powerful graphics processors. CRONOS focuses on understanding how a signal changes over time and flags anything that is different from normal, stable patterns.

Tests on different types of data, like vibrations from machines, body signals, and movement tracking, show that CRONOS can detect many faults without any training. It performs well compared to trained models in several cases. However, very unpredictable situations can still be difficult. In stable environments with plenty of good data, a carefully trained model might still be the most cost-effective choice. CRONOS is best for situations where the lack of labeled data is the main obstacle.

The predictable nature of CRONOS is very useful in industries where rules and safety are important. Unpredictable computer systems can be a problem in these areas. Predictability allows for clear answers about why an alarm went off, how changes were made, and if the system can be used without needing a lot of computing power. These factors are considered important early on, even before trying to get small improvements in accuracy.

The CRONOS project is looking for input from others about where the problem of needing more failure data is common, what workarounds people use, and what would make them trust a system that can detect problems without training in real-world use.

CRONOS uses technologies like AI and is designed to run on platforms such as ARM Cortex-M4, AWS, Azure, and GCP.