Kamil Židek

Model-based Evolution of a Fast Hybrid Fuzzy Adaptive Controller for a Pneumatic Muscle Actuator

Pneumatic artificial muscle-based robotic systems usually necessitate the use of various nonlinear control techniques in order to improve their performance. Their robustness to parameter variation, which is generally difficult to predict, should also be tested. Here a fast hybrid adaptive control is proposed, where a conventional PD controller is placed into the feedforward branch and a fuzzy controller is placed into the adaptation branch. The fuzzy controller compensates for the actions of the PD controller under conditions of inertia moment variation. The fuzzy controller of Takagi-Sugeno type is evolved through a genetic algorithm using the dynamic model of a pneumatic muscle actuator. The results confirm the capability of the designed system to provide robust performance under the conditions of varying inertia.

Experimental Validation of Nominal Model Characteristics for Pneumatic Muscle Actuator

Pneumatic artificial muscles belong to a category of nonconventional pneumatic actuators that are distinctive for their high power/weight ratio, simple construction and low price and maintenance costs. As such, pneumatic artificial muscles represent an alternative type of pneumatic actuator that could replace the traditional ones in certain applications. Due to their specific construction, PAM-based systems have nonlinear characteristics which make it more difficult to design a control system with good performance. In the paper, a gray-box model (basically analytical but with certain experimental parts) of the one degree-of-freedom PAM-based actuator is derived. This model interconnects the description of pneumatic and mechanical part of the system through a set of several nonlinear differential equations and its main purpose is the design of intelligent control system in simulation environment. The model is validated in both open-loop and closed-loop mode using the measurements on real plant and the results confirm that model performance is in good agreement with the performance of real actuator.

Enhanced Dynamic Model of Pneumatic Muscle Actuator with Elman Neural Network

To make effective use of model-based control system design techniques, one needs a good model which captures system’s dynamic properties in the range of interest. Here an analytical model of pneumatic muscle actuator with two pneumatic artificial muscles driving a rotational joint is developed. Use of analytical model makes it possible to retain the physical interpretation of the model and the model is validated using open-loop responses. Since it was considered important to design a robust controller based on this model, the effect of changed moment of inertia (as a representation of uncertain parameter) was taken into account and compared with nominal case. To improve the accuracy of the model, these effects are treated as a disturbance modeled using the recurrent (Elman) neural network. Recurrent neural network was preferred over feedforward type due to its better long-term prediction capabilities well suited for simulation use of the model. The results confirm that this method improves the model performance (tested for five of the measured variables: joint angle, muscle pressures, and muscle forces) while retaining its physical interpretation.

Wireless Device Based on MEMS Sensors and Bluetooth Low Energy (LE/Smart) Technology for Diagnostics of Mechatronic Systems

This paper deals with usability of MEMS sensors for diagnostics of mechatronics system state wirelessly. We can acquire basic kinematics and dynamics mechanism parameters (spatial position, speed, acceleration, vibration, angular rate, orientation, etc.) and some environment condition (local/remote temperature, humidity, pressure, electromagnetic noise) by MEMS sensors. Acquired data are sent to remote application in desktop computer. This system can replace expensive and separate diagnostic tools by small integrated solution with one wireless communication interface (with limitation of MEMS sensors precision). This solution can be battery powered with long operation time, because there is used new wireless technology based on Bluetooth 4 protocol (Low Energy/Smart Bluetooth). Some of integrated MEMS sensors measures same variable on different measuring principle. For example angle can be acquired from three different sensors: magnetometer, accelerometer or gyroscope. Combination of these sensor data can significantly improve value accuracy. The designed diagnostic tool can serve as an inertia measuring unit IMU or Wireless IMU (WIMU).

Recognition of Assembly Parts by Convolutional Neural Networks

The paper describes the experiments with the use of deep neural networks (CNN) for robust identification of assembly parts (screws, nuts) and assembly features (holes), to speed up any assembly process with augmented reality application. The simple image processing tasks with static camera and recognized parts can be handled by standard image processing algorithms (threshold, Hough line/circle detection and contour detection), but the augmented reality devices require dynamic recognition of features detected in various distances and angles. The problem can be solved by deep learning CNN which is robust to orientation, scale and in cases when element is not fully visible. We tested two pretrained CNN models Mobilenet V1 and SSD Fast RCNN Inception V2 SSD extension have been tested to detect exact position. The results obtained were very promising in comparison to standard image processing techniques.

Management of linear quadratic regulator optimal control with full-vehicle control case study

Linear quadratic regulator is a powerful technique for dealing with the control design of any linear and nonlinear system after linearization of the system around an operating point. For small systems, which have fewer state variables, the transformation of the performance index from scalar to matrix form can be straightforward. On the other hand, as the system becomes large with many state variables and controllers, appropriate design and notations should be defined to make it easy to automatically implement the technique for any large system without the need to redesign from scratch every time one requires a new system. The main aim of this article was to deal with this issue. This article shows how to automatically obtain the matrix form of the performance index matrices from the scalar version of the performance index. Control of a full-vehicle in cornering was taken as a case study in this article.

Diagnostics of Surface Errors by Embedded Vision System and its Classification by Machine Learning Algorithms

The Article deals with usability and advantages of embedded vision systems for surface error detection and usability of advanced algorithms, technics and methods from machine learning and artificial intelligence for error classification in machine vision systems. We provide experiments with following classification algorithms: Support Vector Machines (SVM), Random Threes, Gradient Boosted Threes, K-Nearest Neighbor and Normal Bayes Classifier. Next comparison experiment was conducted with multilayer perceptron (MLP), because currently it is very popular for classification in the field of artificial intelligence. These classification approaches are compared by precision, reliability, speed of teaching and algorithm implementation difficulty.

Comparison of Performance of Optimized Varela Immune Controller and PID Controller for Control of Time-Delay Process

Varela immune controller is a kind of nonlinear controller, which is said to have good anti-delay capabilities. We compare the performance of simulated annealing optimized improved Varela immune controller and optimized PID controller for controlling a process with very long time delay (approximation of biomass-fired boiler temperature control). The results confirm that Varela immune controller is indeed capable of stabilizing the process while being very robust even to extreme changes in process parameters (time constant and time delay). In addition to that, it is also found out that properly (optimally) tuned PID controller is capable of achieving similar performance. The problem of controller tuning is relevant for both controllers but there are no tuning rules for immune controllers, which might favor the use of conventional PID controller. On the other hand, Varela controller has greater flexibility due to its more complex structure, which might help to adapt it to some special kinds of processes.

Diagnostics of Errors at Component Surface by Vision Recognition in Production Systems

The article deals with the diagnostics of components surface after painting by camera system in real-time. This solution is especially suitable for implementation to automatized production line above the conveyor belt. The faults on the part surface can be detected as scratches, imperfect surface coverage and dirt stuck to the surface. The scratch detection is based on edge detectors, imperfect coverage are checked by histogram comparison and all other errors are detected by counter detectors. The developed software uses open source library OpenCV and is written in C++ language. The software solution is platform independent. Final algorithm is implemented to embedded device based on SoC.

Diagnostics of Product by Vision System

This article describes the vision system, which is designed for diagnostics of defects in casted products. In the first part an overview about image processing, edge and pattern recognition algorithms and current status in available free and commercial vision libraries is found. For the described task we selected open source Aforge .NET library. The next part describes common defects in casted products. Modular education system MPS 500 from Festo with conveyor and palette with plastic parts is used for simulation of production system. This system contains an industrial robot which can be used for sorting defective parts. The selected vision library is used for two level diagnostics of algorithm implementation. The first level algorithm detects position of part, its dimensions and edge disturbances. The second algorithm detects any defects inside of a part. The basic algorithm is presented only for circular shape with red color texture, but can be easily extended to other basic shapes by shape detector.