Onboard, Real-time Anomaly Detection via 4-th Generation AI

Modern products such as automobiles, aircraft, ships, military equipment, etc., function thanks to a multitude of software, hardware, and sensors/actuators. The amount of data they produce is massive. Their key characteristic is their immense complexity.

Because of technology advances this complexity increases and is becoming a threat of its own. This is because high complexity inevitably implies fragility. Highly complex systems can suddenly fail and in an unexpected manner. As complexity increases this becomes more likely.

Ontonix has developed a real-time monitoring capability for complex products, based on its Artificial Intuition technology. The beauty of the technology lies in the fact that it doesn’t require Machine Learning to identify the presence of an imminent anomaly. Besides, certain anomalies are very rare and learning to recognize them is practically impossible.

The Artificial Intuition-based monitoring system developed by Ontonix produces, in addition to anomaly early warning, information useful for Condition-based Maintenance.

Condition-based Maintenance (CBM) is a maintenance strategy that monitors the real-time condition of an asset to determine what maintenance needs to be performed. Unlike preventive maintenance, which adopts calendar or statistics-based strategies to determine when to schedule maintenance, Condition-based Maintenance dictates that maintenance should only be done when necessary. Not too early, not too late.

The goal of Condition-based Maintenance is to continuously monitor assets to spot impending failure or anomalies, so maintenance can be proactively scheduled before the failure occurs. The idea is that this real-time monitoring will give maintenance teams enough lead time before a failure occurs or asset performance drops below an optimal level.

Fourth generation of Artificial Intelligence is ‘Artificial Intuition,’ and has been in development by Ontonix over the past decade. It enables computers to identify threats and vulnerabilities without the need for training, just as human intuition allows us to make decisions without specifically being instructed on how to do so.

There exist numerous, rapidly changing scenarios in which fast decision making is needed but there are not enough examples from which to learn. Artificial Intuition is able to monitor highly complex systems and identify and signal imminent anomalies. The technology, which does not require any form of Machine Learning, has been tested during the past decade in numerous sectors, such as defense, manufacturing, or medicine. Since generating training sets and using them in a Machine Learning context in unnecessary, Artificial Intuition is extremely fast and ideal for real-time edge-computing or on-board applications.

To make the technology widely and safely available – in automobiles, aircraft, ships, medical equipment, routers, transformers, or military platforms – a new edge-computing capability is offered. The approach is simple: the entire Artificial Intuition-based software system is delivered on a mini PC, which is connected to a USB port on the client’s platform. The current implementation adopts an NiPoGi Ntel Atom x5 z8350 Mini PC, 8 GB DDR3 RAM 128 GB eMMC, 4.2/4K HD/USB 3.0, as shown below:

The functioning is as follows:

When power is turned on, the system starts to function automatically.
A data frame of N samples of each of the M channels one wants to monitor, must be created on the client’s platform and dropped into a specific directory on the USB drive. This can be done once a second, once a minute or with any desired frequency.
The system automatically analyses it and returns a list of current potential fragilities as well as anomaly early-warning triggers, if any. This is done with the above mentioned frequency.
Anomaly early-warnings may be displayed on client’s dashboard or on a separate display, such as the one shown above.
Information for Condition-Based Maintenance is stored on the drive for subsequent use by the manufacturer or operator of a given platform.

An example of the type of message issued by the system is illustrated below.

When an anomaly is imminent, the system indicates the time, the frame number, type of anomaly as well as the key variables (channels) responsible for the anomaly in question.

The system doesn’t require any software installation on client’s platform. There is no need to install specific PC boards, or software. All the necessary computational resources and storage are external to the client’s IT infrastructure. Wifi may be disabled to avoid security issues. All that is required is to establish a two-way I/O data transfer, following a very simple protocol.

Both Linux and MS Windows configurations are available.