The inertia of the shaft is set to zero since its influence is negligible to the gravitational and inertia forces of the elevator cabin.
The third equation describes mechanical dynamics of the motor ( mechanical equilibrium). The factor between both is again, the motor parameter k. The second equation shows the linear relationship between the electromotive force voltage Ui and the rotational speed n(t). The torque is proportional to the motor’s characteristic k. In this article we will develop a simple elevator model which consists of two main parts: the DC-motor armature (electrical part) on the left and the mechanical structure on the right side.įirst equation is the formal relationship between the current i(t) and the driving torque M. This article could be also interesting for you if you used MATLAB during your studies and now consider to switch to Python due to save costs or to use a larger ML/DL stack. Since we use Python for solving the ordinary differential equations (ODE) you should know about creating, manipulating and plotting NumPy arrays.
You should have some basic knowledge in maths, physics, mechanical and electrical engineering. I wrote this article for everyone interested in applying machine learning algorithms on real-world engineering problems. Or you could use this knowledge to build simulation environments as a training camp for your sophisticated, neural network based RL-agents. This article tries to be a motivation for you to take a closer look on concepts introduced in control theory and apply them in your daily life for understanding things, playing around with parameters and observing the behavior of your system to changes you introduce. RL agents will adjust their internal non-linear state-value or state-action function to those specific situations and achieve a superior performance as compared to traditional control methods. Physical models often simplify our non-linear real world in order to allow (numerical) calculations. Alternatively, you could have a sufficiently accurate physical simulation of your environment, which you could use to train the baseline of your agent and later, after you reached a certain threshold with its performance, deploy the agent into real environment to transfer and adjust its knowledge. in the AD example: real traffic with real vehicles)-which can be expensive or even dangerous. Reinforcement learning tries to solve the problem of controlling complex systems, but training agents requires thousands of epochs of trial-and-error (exploration/exploitation) in the target environment (i.e. Sure, Autonomous Driving (AD) stack heavily depends on Deep Learning algorithms for perception using radar, LIDAR and video as input, though the control part is still managed by methods developed over centuries in system theory and control engineering. Surprisingly, there are only few prominent examples for applying machine learning as core element in safety critical systems inside the engineering domain. You hear about groundbreaking advances in language processing, computer vision or reinforcement learning everyday.