We keep up with the security industry

ARTIFICIAL INTELLIGENCE

The times when CCTV systems were only used to monitor the necessary spaces or for perimeter security are long gone; today there are more effective tools available for more efficient and automatic evaluation of scenes. The system uses the potential of artificial intelligence, and we can solve almost any problem by training a deep neural network.

The application uses special deep neural networks for video analysis. Artificial intelligence is applied to video in various resolutions for early detection and recognition of phenomena of interest, from the violation of a perimeter by a person to complex directional detection or analysis of other objects and phenomena.

QUALITY CONTROL

Artificial intelligence is a tool for advanced quality control or machine diagnostics that uses machine vision and deep neural networks; it is perfect for image, video, sound or vibration recognition. Neural networks are taught much like humans with examples (patterns) that they later use to respond to stimuli from the real world. Their advantage is the ability to generalize from the learned knowledge, which they can then apply in conditions different from those under which they were trained.
This allows the system to detect surface defects of materials on static photos or video, where dependencies in the development of the video can also be detected for quality control, selection of parts, or control of production processes. The system is also supplied as a complete unit with cameras, lighting and a computer for these purposes.  2D or 3D measurement of dimensions can also be added to the system.
The primary learning system prior to live deployment follows the patterns that the user indicates in the web browser, thus defining the types of defects to be detected. This core learning takes place off the target production line on a powerful server.
The system can subsequently continue to be taught while in operation by gradually adding new patterns. If a new defect occurs, the user simply marks the new images and lets the system retrain itself automatically. Everything operates on a strictly user basis while the production line is running without the need for programming.

FOUR OUTCOMES OF MACHINE LEARNING AND BENEFITS FOR MANUFACTURERS

1. Predictive failure detection and alarm generation are areas in which companies can save the most. If a production line is to be down, it can be much more cost-effective to shut it down early and perform maintenance than to shut it down in the middle of the production process. Predicting failures and generating alarms based on algorithms can save a lot of money.
2. Process optimisation is another popular area. The system generally provides recommendations for setting setpoints and regulated quantities for systems. There are two main types of process optimisation: open-loop and closed-loop. Open-loop optimisation involves interaction with the user, where the system can recommend changes to optimise the process, and a technician or other expert reviews the recommendations and decides whether to implement them. Closed-loop optimisation completely eliminates human intervention and tuning recommendations are applied automatically.
3. Anomaly detection detects deviations from normal operating conditions. This can show you when a process is not running optimally, or predict bad a poor production process or equipment errors. These systems typically provide a number that indicates how close the process is to normal operating conditions. The results can be thought of as another “sensor”, and the results of the anomaly detection algorithm can be used to do anything from alerting you to an abnormal condition to stopping a process. Although anomaly detection is similar to predictive failure detection and alarm generation, it shows you how a process is performing right now rather than predicting how it will perform in the future.
4. Defects are usually analysed by image recognition algorithms; this analysis can be very useful for classifying parts and detecting abnormalities.

SAFETY CONTROL

We can now increase the safety of workers in three internally tested scenarios. We are gradually deploying these solutions in pilot projects in industrial companies.

• Machines stop before an accident happens. 

Our system can distinguish between the human body and processed materials, other machines, etc. If a human hand approaches a part of the machine that could cause an injury, we stop the machine. This is our original idea, which is why we completed this feature first. Conversations in industrial companies repeatedly pointed us to two other scenarios:

• Authorisation of workers to operate equipment. 

The software can recognize individuals and allow (or not allow) them to operate a machine, or it can only make some of its functions available to certain people. Two companies voiced this requirement. One company wanted to only enable certified workers to operate a forklift, and the second company wanted their CNC machine to allow the service technician to open it while it is running and adjust it after his authorisation. Both scenarios naturally have multiple potential solutions. We were told, however, that the simplest solution, namely locking the machine with a key, cannot be realistically used, as employees usually leave the keys in the machine for practical reasons.

• Inspection of personal protective equipment 

– safety helmets, vests, facemasks, etc.; in this case, the machine can only be started by a worker who is properly equipped, but we were asked for something much simpler: they requested that the machine check whether night shift workers are wearing protective equipment (when the managers aren’t present) and record violations.