Identification and the Human Factor

Since the early 1900s, the desire to organize work tasks on the production floor in order to obtain more efficient processes has continued to evolve. Mass production lines with repetitive tasks and cyclical routines were perfected to a degree where industrial ethics began to imply mechanisms in which man was not so exposed to such aggressive exploitation.

All we need to do is recall the 1936 film Modern Times, where Charles Chaplin plays a worker immersed in the economic crisis of the Great Depression and the misfortune that accompanies him as a result of not being able to adapt to an assembly line. Almost a century later, the specter of man's displacement by automation technology continues to prowl between industrial advances and economic challenges.

Along with Fordism, a chain production system, lines of thought emerged around the improvement of production, such as Taylorism, which, although it was heavily criticized because it generated high repetitiveness and monotony, as well as a total detachment from the human part of the operator, reinitiated the search for operational efficiency, under a scientific approach, the increase of control and processes management, and wider company profit margins.

Plant times and movements, as a key concern, continue to be the center of attention and are gaining strength with the arrival of information technologies (IT) in manufacturing and its pairing with deep learning (DL) and artificial intelligence (AI) tools. Taylorist principles, such as quantitative analysis of work, training and the correct selection of personnel for specific tasks, the alignment of tasks and interaction between operators and the production team, as well as the specialization of production techniques, will be better refined, but rescue the human approach that organizations such as OSHA are advocating for.

The development of the Industry 4.0 production ecosystem or the Industrial Internet of Things (IIoT) brings with it vision and sensory equipment that is beginning to be applied in the operational management of the production floor. It should be noted that the application of these components in assembly or material handling lines is not new, and they are a great help in identifying and classifying material in production. Shapes, sizes, colors, volume and other indicators are used to ensure continuity, prevent errors, and prevent potential hazards and accidents, among other applications.

When applied in the recognition of workspaces, data related to time and movement can also be obtained or the activities of operators outlined. Sensors and cameras collect information on movement, proximity, acceleration and orientation. Audio and video complement the reading and fill databases from which parameters are extracted that can then be used in management systems.
The recorded activities can warn of falls, bad posture and even identify from how a person walks if they are tiredness, bored — movements that are not suitable for an operation.

When used in ergonomic areas, a space can be redesigned to improve an operator's working conditions, determine if he requires the support of another operator, equipment or tool, and even provide training if skill deficiencies are identified.

Vision systems have greatly thrived on the basis of body and facial identification models used in AI models. Such models have been widely developed in areas such as autonomous car driving because they are capable of detecting gestures and movements that, in those cases, determine if a driver is tired, too distracted by devices such as cellphones or even if his mind could be wandering rather than concentrating on driving the vehicle. The algorithms of such systems, in communication with other sensors in the vehicle, will launch alerts or take control of the car to prevent accidents.

In a production cell, this information will be used in learning models based on DL technology to determine new training and training routines, assess an operator's skills and thus reduce learning curves. A training program can be designed according to each function and task, or to each machine or device with which the person will interact. It will even be possible to determine the ideal time in a working day to maintain productivity without affecting quality (another of the most severe criticisms against Taylorism).

Using DL technology, complex data models can be accessed from convolutional neural network (CNN) systems, which will provide visual information and facilitate the understanding and design of new work cell layouts, routines and operational tasks or training programs. Academic experiments have already been carried out around this in manufacturing activities, in maintenance operations and in the production of cars and bicycles, calculating working times and improving quality levels in the operation.

The use of vision and sensor technologies, in connection with AI models, will allow traceability schemes to be established in workflows, even allowing virtual operation processes to be implemented to improve factual operations.