Cookie Consent by Free Privacy Policy Generator - Průmyslová bezpečnost
May 22nd, 2024
4 min read

Innovation of machine safety systems based on artificial intelligence

How does a neural network assess the risk of injury in the workplace?
The purpose of machine safety systems is to reduce the risk of injury or death to machine operators to an acceptable level. Various technologies are used to do this , such as barriers , optical barriers and so on. These systems often cause unwanted machine stops, e.g. a sheet of flying paper breaks an optical barrier and subsequently the machine or the entire production line is stopped. This results in downtime and subsequent production losses with an adverse economic impact on the company.

These shortcomings are eliminated by machine vision-based systems using a neural network. This article shows one way to improve the reliability of neural network output prediction using mathematical reliability theory. An important aspect of the technology using AI is also the simplicity of implementation, which will allow its deployment even in small and medium-sized enterprises.

Any modern machine or production line must be designed to minimise the risk of injury or death to the operator. This is done using the principles of functional safety, where an analysis of the risks associated with the operation of the machine is carried out at the outset. These risks can arise from hardware failure, software error or even human factors. On the basis of the assessment of these risks, sufficiently robust measures must be designed to reduce the risks to an acceptable level, as shown in (EN 61508-5 ). A schematic risk reduction procedure is shown in Fig. 1. It should be remembered that the obligation to reduce the risk sufficiently is both on the part of the machine manufacturer and on the part of the employer, as stated in (EU 2006/42/EC).

Schematic risk reduction process

- An EUC (Equipment Under Control) is a controlled system where a failure (defect) creates a risk of personal injury, death
- SIS (Safety Instrumented Systems) is a safety instrumented system that reduces risk through safety functions.

It is therefore a 'safety - related system' and it would be good to have an idea of its safety performance - reliability towards the main goal of the deployment of the device, i.e. the protection of people in the guarded perimeter of the machine or technology.
The output of the neural network decision-making process is the criteria that form a complete set of phenomena. These are the following phenomena:

Phenomena as outputs of the neural network decision process

Various metrics can be used to evaluate the neural network and its level of training. There is a need to have a set of validation images where some of them are human and some are not.
The basic metric is called ACCURACY and is defined as the number of correct predictions on the validation set of images. This metric is fundamentally influenced by the use of an appropriate neural network model and the degree to which it is trained, and assesses the degree of reliability of the neural network prediction.
Another metric used for neural network evaluation is RECALL. It is determined by selecting images from the validation set of images where a human is actually present in the guarded space of the machine.
The third metric used is PRECISION, which indicates the confidence level if the neural network predicts the presence of a human in the guarded machine space.
In the neural network evaluation, we further define the so-called Threshold. This parameter defines the minimum probability threshold with which a human has been positively identified in the guarded machine space. For example, if this level is set to 0.7, the presence of a human will only be predicted if he or she is identified with a probability of 0.7 or greater. An example of a human identification probability is shown in Fig.3.

Example of human identification probability

As the value of Threshold parameter increases, the number of false positive identifications will decrease.
The above uses Invanta's highly innovative machine vision-based technology. The digital camera signal is fed onto the field bus of a dedicated processor running a neural network. This evaluates the area being monitored and activates the security system when a person or even just a body part is detected. As in any workplace, the basic safety rules and principles for the site must be followed.

Aktuelle Artikel

Alle anzeigen