Frank W. Paul

Vision guided robotic fabric manipulation for apparel manufacturing

Methods are described for vision-guided robotic control of fabric motion for performing simulated joining operations for apparel manufacturing. The determination of robot motion paths is based on visual information defining the position of the fabric edges in world coordinates. The usefulness of the shape analysis and motion control algorithms is demonstrated by experimentation with an integrated robot and vision fabric-handling system. Results of these tests show that using machine vision for planning robot motion provides an effective solution for implementing automated robotic fabric manipulation.<<ETX>>This paper details methods pertaining to the vision guided robotic control of fabric motion for performing simulated joining operations along the fabric edge. Robot motion paths are determined from visual information defining the position of the fabric edges. The function of the shape analysis and motion control algorithms is demonstrated via experimentation with an automated fabric handling system, Results of these tests show that using machine vision for planning robot motion provides an effective solution for implementing robotic fabric manipulation.

Multiple neural-network-based adaptive controller using orthonormal activation function neural networks

A direct adaptive control scheme is developed using orthonormal activation function-based neural networks (OAFNNs) for trajectory tracking control of a class of nonlinear systems. Multiple OAFNNs are employed in these controllers for feedforward compensation of unknown system dynamics. Choice of multiple OAFNNs allows a reduction in overall network size reducing the computational requirements. The network weights are tuned on-line, in real time. The overall stability of the system and the neural networks is guaranteed using Lyapunov analysis. The developed neural controllers are evaluated experimentally and the experimental results are shown to support theoretical analysis. The effects of network parameters on system performance are experimentally evaluated and are presented in this research. The superior learning capability of OAFNNs is demonstrated through experimental results. The OAFNNs were able to model the true nature of the nonlinear system dynamics characteristics for a rolling-sliding contact as well as for stiction.

Acquisition, placement, and folding of fabric materials

This paper discusses the technical issues associated with the acquisition, placement, and folding of fabric materials with mechatronic devices and machines. Earlier work in this area considered the acquisition of fabrics from a stack of materials. Numerous techniques were evaluated and suggested as a satisfactory way to provide “pick and placement” of such materials for various automated processes. Several end‐effector devices were developed which used “pinching and stretching” and “multiple roller” approach for handling, placing, and smoothing fabrics on a flat surface. Fabric wrinkles were detected using a feedback laser sensor to assist in the placement and positioning of fabrics. Later work focused on the positioning of fabrics that required issues of alignment, placement and folding for a variety of fabric operations. A two‐dimensional process considered precise placement, laying down, and then folding of a fabric material to have matched ends using a robot manipulator using visual feedback sensing. Additionally, three‐dimensional diagonal folding of fabric was considered based on knowledge developed from the two‐dimensional case. Work was also conducted to mathematically model and measure deformations of limp fabrics and how wrinkling influenced the process of fabric smoothing, alignment, and folding. The results of this work showed that different fabric types (lighter versus heavier) have different sensitivities and hysteritic effects with respect to wrinkling, smoothing, alignment, and folding.

Real time adaptive control experiments with a multiple neural network based DCAL controller

The results of real time experiments are presented for a desired compensation adaptive law (DCAL) neural controller developed using orthonormal activation function-based neural networks (OAFNN). The task of the controller is to track the desired trajectory for a class of nonlinear systems. Multiple OAFNNs are employed in the neural adaptive controller for feed-forward compensation of unknown system dynamics. The feed-forward compensation is based on desired trajectory allowing flexibility of its off-line computation. Choice of multiple OAFNNs allows a reduction in overall network size reducing the computational requirements for real time implementation. The network weights are tuned online, in real time. The overall stability of the system and the neural networks is guaranteed using Lyapunov analysis. The effects of network parameters on system performance are evaluated and presented in this research. The superior learning capability of OAFNNs is demonstrated through experimental results.

Multiple neural network based DCAL controllers using orthonormal activation function neural networks

Direct adaptive control schemes are developed using orthonormal activation function based neural networks (OAFNNs) for trajectory tracking control of a class of nonlinear systems. Multiple OAFNNs are employed in these controllers for feedforward compensation of unknown system dynamics. Choice of multiple OAFNNs allows a reduction in overall network size reducing the computational requirements. The network weights are tuned online in real time. The overall stability of the system and the neural networks is guaranteed using Lyapunov analysis. The developed neural controllers are evaluated experimentally and the experimental results support the theoretical analysis.

A Neural Network Spectrum Analyzer

The Fourier Series Neural Network (FSNN) is used as a Neural Network Spectrum Analyzer (NNSA) to gain the advantages of computational parallelism, self-learning intelligence and the capability of modeling complex systems with multiple variables. A comprehensive comparison study between the NNSA and a traditional spectrum analyzer based on the Fast Fourier Transformation (FFT) was conducted to evaluate the performance of the NNSA when used for transfer function identification. This evaluation was carried out using computer simulations and experimentation in which a mechatronic robot end-effector was used as the plant to be identified using the two methods. The identification results described in this paper demonstrate that the NNSA is able to approach the same model accuracy as FFT.

Characterization and Control of Object Acquisition with a Sensored Robot Hand

This paper presents the thoughts, results and on-going research directed at the characterization, understanding and control of object acquisition using a robot positioned sensored-based hand. The paper overviews the current literature relating to robot hand design, and addresses the questions of: * object characterization, * principles which govern object acquisition, * control design considerations for object acquisition. Preliminary results are presented which show forces associated with uncontrolled object acquisition, finger and object material influence on the transient acquisition forces, and a design for velocity and position control during object acquisition.

A hierarchical system for robot-assisted shirt collar processing

A robotic workstation which is controlled by PC-level computers for performing complex assembly tasks on shirt collar components is discussed. Assembly procedures and equipment designed to aid the integration of robotics and visual sensing with the tasks at hand are discussed. A hierarchical control scheme where planning is achieved through a man-machine interface with intelligent software is discussed

Tactile Sensors: Application Assessment for Robotic Handling of Limp Materials

Tactile sensors have a potential role in assisting the robotic manipulation of limp materials such as apparel fabrics, leathers, flexible composites, thin metals (foils), and polymer sheets. A review of the methods and principles utilized in existing tactile sensor technology is presented coupled with concepts for integrating tactile information with robot control. Possible uses of tactile sensors for limp material handling with robot manipulators are suggested. Research areas where further exploration and development is needed are identified.

Nonlinear System Identification with Fourier Series Neural Networks

Abstract This paper addresses the issues concerning the Fourier Series Neural Network (FSNN) identification of nonlinear dynamic systems as well as the Boolean functions. The behavior of the FSNN identifier was evaluated through its application to the modeling of the systems which can be characterized by the describing functions, difference equations and XOR function. The results of the evaluation are described and analyzed in the paper.