Artificial intelligence (AI) is steadily being used throughout industries. While most people might think of it as robots, or high risk only, there are many benefits to AI for both companies and individuals. A few examples of the uses of AI as solutions to industrial challenges are quality control, catastrophic failure predictions, and efficiency. There are two types of AI: narrow AI (weak AI) and general AI (e.g., AGI and strong AI). Narrow AI refers to AI that is capable of specific, individual tasks, such as facial recognition, chess playing, searches (SIRI), playing music (ALEXA) and driving a car. AGI refers to AI that has cognitive abilities, such as the ability to learn methods and apply data to algorithms to learn and gain in capability across a broad range of tasks.
Another laudable feature of artificial intelligence is chatbot development. Chatbot development using NLP enables chatbots to understand customer requests and give relevant responses.
Risks are associated with any use of technology, such as the possibility of the AI to crash, or become hacked. In the case of war, there is a race currently on-going to develop AI technology capable of taking the place of a soldier manning a weapon. The “slaughterbots” are lethal autonomous weapons systems, that are being developed to be able to target and kill subjects that fit their algorithms (e.g., facial recognition), without human intervention. AI in this form does present a possible solution to times of war, as human lives used (on the side controlling the AI weaponry) would be decreased. However, this technology is also difficult to shut down since you don’t want people to be able to maliciously disable it. Therefore, this creates a risk of being unable to disable the AI weapons system if the AI’s goals are not aligning with the goals of those using it, such as recognizing friend from foe.
These goals are based upon the programming installed and algorithms provided to create conclusions from data gathered. AI is used every day without people realizing it, but it could have a greater impact on the average person’s life. In a study at a Frito Lays company, AI was used to track the positions that people were in as they completed their tasks at work. The data was captured and run through algorithms, comparing the movements and postures of workers to information programmed concerning what postures are high-risk for injuries. This gave the workers alerts when they were using high risk postures while performing tasks and sent the data to management. They were able to collectively reconfigure how to perform tasks in ways that reduced injuries, as well as management to reconfigure the spatial arrangement for more conducive navigability for the workers. After this was implemented, there was a nineteen percent reduction in recordable injuries as well as a sixty-seven percent reduction in loss of worktime.
AI could also be used in such an instance to create the solutions without the intervention of people. However, this technology application can be used for a broader range of purposes than just workers in an industry. Average people, at home, could use it to monitor their own postures to improve their overall health and reduce the likelihood of injury. In the industrial setting, there are many issues that arise, such as quality control and machine down-time.
Regarding quality control, few AI technologies are being utilized. Primarily, the use of cameras and visual inspection is the method of quality control. However, by using AI to check every product manufactured, there would be fewer defective products sent to customers which creates a cost savings for the company to address issues without having to reimburse customers for costs expended.
Features can be programmed into AI, such as the color and shape of ripe fruit, as one method of quality control. Deep neural network (DNN) can also be used, as it is a learning algorithm in which data is collected and taught to the AI. This is also referred to as deep learning (DL). Lifelong continual learning DNNs (L-DNN) would be the ideal next step to decrease the time it takes for an AI to learn what is acceptable for quality control. The L-DNN provider would have generic features and information (such as curvature, properties, edges) that all AI can instantly download and add to their database. This way, the individual manufacturers would only have to give them a few images of acceptable products for the AI to develop a baseline of quality control standards to implement.
Currently, some industries utilize machine learning (ML) and DL, subparts of AI. However, the integration of full AI capabilities with the ML and occasion DL used in Prognostic & Health Management (PHM) systems would increase the effectiveness of its predictive maintenance (PM). A PMH system is a field of engineering where a machine and its components health conditions are analyzed with statistic and physics. The system is then able to detect faults in equipment and determine expected time to failure based upon these faults and historical data that has been inputted or gathered by the system. An increased effectiveness of a PM system causes less catastrophic failures, which in turn creates more machine up-time and efficiency (due to planned shutdowns versus products being ruined when machines unexpectedly fail).
In addition to a PMH system, scheduled maintenance is generally used, which is a narrow AI. If L- DNNs AI are implemented for the use of PM, then they would be capable of learning when machines, equipment and components should have scheduled maintenance based upon the information and data it has obtained, being input into the AIs algorithms. If manufacturers have any PM system in place, it generally only consists of sensors and future operations data, allowing it to have the fault detection system. A study was conducted across several companies over the change before and after implementing a PMH system. Sixty percent of the companies involved reported an average of a nine percent increase in machine uptime. Based upon this study, a full AI based system would prove to be beneficial. The AI could use the data gathered from anomalies detected during machine operations and create a cause correlation based upon the conditions that were present when the anomaly occurred. The data necessary for this correlation would consist of variables such as signals (voltage, temperature), dynamics measured and sample value discretization.
Artificial neural networks (ANN), a part of DL, and support vector machines (SVM) would be necessary to find optimal hyperplanes based upon this data. Also, if this is merged with physics based upon data-driven damage models, then the AI could decipher what is an acceptable anomaly versus what is a sign of an imminent failure. One hold-up of the implementation of a full AI system is the complexity of the mathematical equations necessary to account for all possible outcomes that the AI might need to compute.
Based upon the information discovered concerning the current uses of artificial intelligence as well the possibilities of future artificial intelligence implementations, it seems logical that AI can solve some current industrial challenges, with minimal risk. There are inherent risks of AI taking the place of humans, where humans have the potential to be injured by technological malfunctions or missing algorithms. Future study and safety implementations are necessary for AGIs to take the cognitive place of humans, as well as narrow AIs to be incorporated into technology for individuals to use daily. If a rudimentary implementation of a PMH system can grant a nine percent average of machine uptime, imagine what a full AI system can create in terms of productiveness, efficiency, and reduction of catastrophic failures!
Top comments (0)