Artificial Intelligence / Machine Learning in Mechanical Engineering Projects - Motivation Booster

With 10 years of experience in machine analysis, patent inventions, project management and business development, I know that Mechanical Engineer (ME) is an excellent program for the mechatronics product development manager. They are good at integrating controls and sensors into their equipment and have a multidisciplinary team to keep everything up to specification, on time and cost-effectively. However, most MEs are helpless when it comes to Artificial Intelligence (AI)/Machine Learning (ML). AI/ML sounds like rocket science to some legacy ME developers and business developers. The purpose of this article is to discuss why this should not be done and to contribute to further research on apatite. We start with a simple definition of Artificial Intelligence/Machine Learning in Mechanical Engineering - "Improving machine performance using real-time data from efficient decision making in uncertain environments" by Boaz Eidelberg. Example - Maximum turbine vibration in a power plant. Should it stop for maintenance or slow down and keep powering up? Chapter The answer to this question depends on real time data such as:

Manipulated Variables

Vibration Amplitude
Vibration Frequency
Source and Shaft Temperatures
Inlet and Outlet Pressures
Steam mass flow rate and more

Environmental Variables:

Floor Vibration
Ambient Temperature, etc.

Design Parametres:

Shaft diameter, length
Number of teeth
Shaft inertia etc.

Data transmission requires sensors and communication, and I am aware of them. ME also understands other mechatronic devices such as PLCs, motion controllers and software. But uncertainty begins with the best decision-making process that requires a demonstration. These are pretty vague. The combination of all these forms the subject of artificial intelligence/machine learning artificial intelligence/machine learning mechatronics. It is essentially a machine designed by ME, seeking excellence in its design, performance and environment. In our example we consider we have sensors that send signal values of job and environment changes to the controller in real time.

The next question in the decision-making process is: What is the expected return if the machine underperforms compared to what would be expected to be paid if the machine were taken out of service, given all the information we know at the moment? If we can calculate the total return for each option, the correct decision would be to choose the option with the highest total return. To calculate the aggregate demand for the "Continue" decision, we need to know the probability that the machine will resume normal operation and multiply this by the electricity generated. This is the expectation of this option. Next, we need to subtract the expected loss, which is the product of the probability of the machine not working (1 - probability of normal operation of the machine) times the price reduction and repair. These are all the expected benefits of choosing to run the machine. Now we need to compare this with all the expected results when the machine is stopped for repair. This equals the fixed price plus 1 probability lower price. The mystery in equation "total increase in demand" (also known as the all-von Neumann -Morgenstern utility function) is the result of a machine failing to supply its operating system. , the environment and the production of benefits. The next question is: What is this going on? To answer these questions, we need a model that relates the consequences of turbine failure to its operation, environment and design. This is where machine learning comes into play. We write a Neural Network (NN) model as shown above, the input on the left is a vector of some function, environment and design, and the output on the right is the result, in our case the failure result. the turbine is still running. There is a connection between the input and output in the neural network, and the weight of the transformation process is added by doing matrix and vector multiplications (called forward propagation) to generate the product prediction. These weights increase as you update the model with more instances of known patterns and known values. We will try to explain this in our next article.

For example, in a power station known in the world, the generator does not work (i.e. it stopped working with result 1) and a certain different ideas appear in the design. We can now run our neural network models in the cloud (eg. Use AWS, IBM, Google or MS Azure, most of which are free). This cloud can feed and process millions or billions of IoT signals in real time, calculate the estimated probability of failure produced by our NN model and compare it to one that doesn't actually work (eg 1). Initially, all NN weights are estimates, so our estimates will be large. However, after each ML run, the error gradient is calculated using one of the many standard AI/ML languages in the cloud (such as Python and Tensor Flow) and the NN weights are shifted in the decreasing direction. Mistake. error, known as the backpropagation process, in which large error gradients will cause large energy changes. In this way, after many runs, the model should converge to the true representation of turbine failure occurring as a function of operating variables, environmental conditions, and design. As steam turbines are used in many power plants around the world, the integration of data from different sources can help generate reliable data for the design of the application. Similar to this example, an energy-intensive product like static recovery can use AI/ML resources and target performance like shorter recovery time with less pain. Therefore, the opportunity for AI/ML innovation in mechatronic manufacturing is endless. As engineers we can be responsible for leading the innovation of mechatronic solutions to real world problems, we don't need to be software or communications technical experts or have advanced training in electrical engineering. We need to have a good understanding of the machine we are building, including the relationship between the machine's intended use and its design, the different processes and environmental changes, design, simulation, prototype testing, and real machine testing.

We must be able to define operational goals for performance such as maximum productivity, minimum downtime, maximum accuracy, and deliver the results we need for AI/ML teams to develop AI/ML models. A specialist knows many options. Then, we have a good chance of completing the AI/ML mechatronics project with a very good team.