Search
  • mikemohseni70

AI and manufacturing reliability: analogy with an autonomous car, military, and medical surgery

By Mike Mohseni




Discussions surrounding AI and machine learning in manufacturing usually revolve around predictive models to assist management with decision-making such as predicting maintenance time or estimating lead time. Digital twins are another attractive area in industrial automation wherein AI provides a fast-performing replacement for computationally intensive physical models for process design or control purposes.


The application of AI in in-line quality control has grown in interest in recent years. In these applications, manufactured parts go through an inspection line where the AI-powered software detects potential defects based on sensor data. The goal is to avoid defective parts escaping the production line and ultimately creating a more reliable manufacturing line.


The following discusses AI’s potential in real-time monitoring of operation quality as opposed to post-process defect detection. In this application, the AI brain continuously monitors the process and detects abnormalities. The potential business cases can be better understood with examples from different industry verticals.


Autonomous cars use AI models for fast and robust assessments of their environment and operate based on feedback. This AI module is part of the car’s operating system and may require external performance assessments. Here, the question is if system reliability is enhanced by using a third-party AI module that independently monitors the performance. This module should be completely segregated from the operating components of the car.


NATO's Innovation hub previously posted a challenge seeking solutions to enhance the reliability of automated maritime robots. It turns out, the leaders and decision-makers were often not convinced to deploy automated maritime robots in sensitive operations that dealt with human lives. The potential case for AI could be an automated control module that monitors the performance of the robot. Given the robot is AI-controlled in the first place, can this redundant AI module convince stakeholders about the system’s reliability in sensitive applications?


The complexity of the automated systems also challenges reliability. Researchers at the Canadian Surgical Technologies & Advanced Robotics found that an add-on vision module can reduce errors in robotic surgery applications. The positional readings from each robot joint inherit a low error. But, robotic arms consist of multiple joints, causing an accumulation of errors and inaccurate computations. The researchers used a modular camera to independently estimate the position of the tool for more robust and precise readings.


In mission-critical manufacturing operations such as structural welding, un-quantifiable reliability or high complexity translate into additional post-process inspections. In these applications, the consequences, and the cost of liability when defects escape the manufacturing line is substantial. The OpEx associated with the inspection downtime and the risk of finding defects after the process can be significant in the metal fabrication industry that runs on razor-thin margins. On the other hand, any decision to boost reliability by replacing machinery usually becomes entangled with high CapEx and complex decision-making processes.


Through understanding the various risks, and the concerns around automated operations in different industries, it can be concluded that an intelligent inspection module to monitor mission-critical manufacturing operations boosts reliability and saves manufacturers significant operational and capital expenditure.