David J. Hoelzle

Basis Task Approach to Iterative Learning Control With Applications to Micro-Robotic Deposition

Two basic requirements in iterative learning control (ILC), among others, are trial-to-trial trajectory invariance and system dynamics invariance. The ILC algorithm must be reinitiated if either the trajectory or system dynamics vary in-between trials. Here, we introduce a framework that flexibly applies ILC such that trajectory and system dynamics constraints are alleviated. The framework exploits the characteristic that many manufacturing operations are comprised of a set of repeated tasks, termed basis tasks here. Instead of applying ILC to the complete operation trajectory, the correct input signal to accurately perform each constitutive basis task is identified by ILC in a training routine. After basis task training, the corresponding input signals, termed basis signals, are applied in a coordinated manner based on instructions from task-oriented machine languages, such as the pervasive G-Code. This framework allows the reference trajectory to be arbitrarily chosen, provided it is comprised of the defined basis tasks. Key definitions, assumptions, and a general application with performance bounds are detailed. An example application on a micro-robotic deposition (μRD) rapid prototyping system displays the utility of the framework in fabricating two distinct structures without reinitiating the ILC algorithm in-between manufacturing operations.

A multimaterial electrohydrodynamic jet (E-jet) printing system

Electrohydrodynamic jet (E-jet) printing has emerged as a high-resolution alternative to other forms of direct solution-based fabrication approaches, such as ink-jet printing. This paper discusses the design, integration and operation of a unique E-jet printing platform. The uniqueness lies in the ability to utilize multiple materials in the same overall print-head, thereby enabling increased degrees of heterogeneous integration of different functionalities on a single substrate. By utilizing multiple individual print-heads, with a carrousel indexing among them, increased material flexibility is achieved. The hardware design and system operation for a relatively inexpensive system are developed and presented. Crossover interconnects and multiple fluorescent tagged proteins, demonstrating printed electronics and biological sensing applications, respectively.

Cross-coupled iterative learning control of systems with dissimilar dynamics: design and implementation

Cross-coupled iterative learning control has previously been applied to contour tracking problems with planar manufacturing robots in which both axes can be characterised as similar systems; having similar dynamics and identical hardware. However, there are many repetitive applications in which dynamically dissimilar systems cooperate to pursue a primary performance objective. This article introduces a novel framework to couple dynamically dissimilar systems while applying iterative learning control, showing the ability to noncausally compensate for a slow system with a fast system. In this framework, performance requirements for a primary objective can more readily be achieved by emphasising an underutilised fast system instead of straining a less-capable slow system. The controller is applied to a micro-robotic deposition manufacturing system to coordinate a slow extrusion system axis and a fast positioning system axis to pursue the primary performance objective, dimensional accuracy of a fabricated part. Experimental results show a 14% improvement in fabrication-dimensional accuracy with only marginal changes in actuator effort, as compared to independently controlled axes.

Micro-robotic deposition guidelines by a design of experiments approach to maximize fabrication reliability for the bone scaffold application

This work aims to facilitate the transition of micro-robotic deposition (microRD) technology from the research bench to a mass manufacturing environment. The bone scaffolding application is targeted; however, the evaluation process developed is applicable to multiple colloidal material systems, length scales, and structure architectures. A design of experiments (DoE) approach is used to develop statistical correlations between three manufacturing treatments (material calcination time, nozzle size, and deposition speed) and defined reliability metrics. All three selected treatments have a significant effect on structure quality. A longer material calcination time improves the deposition of internal features. Logically, a larger nozzle size decreases structural defects. However, an unexpected result is revealed by this study. Higher deposition speeds are shown to either significantly improve or have no effect on structure quality, permitting a decrease in manufacturing time without adverse consequences.

Bumpless Transfer Filter for Exogenous Feedforward Signals

This brief presents a bumpless transfer filter to attenuate signal magnitude bumps at transitions between exogenous feedforward (FF) signals in a feedback/FF control architecture. Importantly, this filter does not modify the carefully constructed exogenous FF signal at time indices away from a signal transition. Filter architecture, stability criteria, and performance considerations are detailed. An example application demonstrates filter efficacy with a piecewise iterative learning control scheme applied to a serial positioning robot. The bumpless transfer filter attenuates exogenous signal magnitude bumps at transitions between signal segments; a smoothed exogenous signal yields improved performance at transition locations and thereby decreases the 2-norm of the error signal by 12% on average.

Iterative Learning Control for robotic deposition using machine vision

This work presents a new application of iterative learning control (ILC) in two respects. Firstly, the output signal is generated by a machine vision system. Secondly, ILC is applied to the extrusion process in micro robotic deposition (muRD), directly addressing the end product quality instead of contributors to end product quality such as position tracking. A P-type and model inversion learning function are both applied to the extrusion process, a system that has nonlinear dynamics and no readily available volumetric flowrate sensor. Theoretical and experimental results show that the nominal system is first order with a pure time delay. Both P-type and model inversion ILC improve the dynamics, with both systems providing better reference tracking. The ILC compensates for the unmodeled nonlinearities, realizing a reduction of RMS error to less than 20% of the initial value for the model inversion approach. Experiments are performed, displaying the ability to extrude precise and seamless closed shapes with the model inversion ILC. This is a necessary requirement for transitioning materials and embedding sensors in multi- material muRD.

A new spatial Iterative Learning Control approach for improved micro-Additive Manufacturing

Iterative Learning Control (ILC) is an effective control algorithm for improving tracking performance in stable or stabilizable systems that track a repetitive trajectory in time. For systems designed to track a position reference in time, there is a natural map between the temporal and spatial domains and researchers have exploited this map to develop spatial adaptations of ILC. However, there are systems in which the spatial coordinate does not have a unique mapping between time and space, such as additive manufacturing systems utilizing raster trajectories. New methods must be developed for these systems. We present a novel reformulation of ILC that is completely based on spatial coordinates. Two-dimensional convolution, as compared to one-dimensional convolution employed in temporal ILC, is applied to innately inform the algorithm the spatial proximity of measured data points. We show that the algorithm can be rewritten as a standard lifted-domain ILC update law, however with an embedded spatial map. Simulations incorporating a model of material ejection in a micro-Additive Manufacturing system demonstrate spatial ILC efficacy.

A Microfluidic Technique to Probe Cell Deformability

Here we detail the design, fabrication, and use of a microfluidic device to evaluate the deformability of a large number of individual cells in an efficient manner. Typically, data for ~10(2) cells can be acquired within a 1 hr experiment. An automated image analysis program enables efficient post-experiment analysis of image data, enabling processing to be complete within a few hours. Our device geometry is unique in that cells must deform through a series of micron-scale constrictions, thereby enabling the initial deformation and time-dependent relaxation of individual cells to be assayed. The applicability of this method to human promyelocytic leukemia (HL-60) cells is demonstrated. Driving cells to deform through micron-scale constrictions using pressure-driven flow, we observe that human promyelocytic (HL-60) cells momentarily occlude the first constriction for a median time of 9.3 msec before passaging more quickly through the subsequent constrictions with a median transit time of 4.0 msec per constriction. By contrast, all-trans retinoic acid-treated (neutrophil-type) HL-60 cells occlude the first constriction for only 4.3 msec before passaging through the subsequent constrictions with a median transit time of 3.3 msec. This method can provide insight into the viscoelastic nature of cells, and ultimately reveal the molecular origins of this behavior.

Bumpless transfer for a flexible adaptation of Iterative Learning Control

This work builds upon a framework for improving trajectory flexibility in systems controlled by Iterative Learning Control (ILC). Here we focus on positioning systems, decomposing a class of trajectories into motion primitives, termed basis tasks. The correct input signal for each basis task is identified in a training routine with ILC. The main development of this paper is a framework to intelligently apply these basis task specific input signals using an adaptation of bumpless transfer techniques. Bumpless transfer is reoriented to seamlessly transition between open-loop ILC signals without attenuating signal content away from the transition points. Experimental results display the effectiveness of the proposed approach on a serial planar positioning robot. Two conditions on basis task sequencing are tested. One which satisfies constraints imposed by previous work, and a relaxed trajectory constraint case designed to further explore trajectory flexibility. Bumpless transfer recovers some of the performance lost by constraint relaxation.

On Spatial Iterative Learning Control via 2-D Convolution: Stability Analysis and Computational Efficiency

Iterative learning control (ILC) is an effective control strategy for improving control performance in stable or stabilizable systems that track a repetitive trajectory in time. The ILC paradigm has previously been extended to the spatial domain; however, spatial ILC (SILC) methods have most commonly been applied to problems in which there is a natural bijective map between the temporal and spatial domains to simply redefine processes in space instead of time. Yet, there are applications in which there does not exist a unique mapping between time and space, such as additive manufacturing (AM) systems utilizing a raster trajectory. In this exploratory work, we present a novel reformulation of ILC that is derived from 2-D convolution in spatial coordinates, compared with 1-D convolution employed in temporal ILC, which innately informs the algorithm of the spatial proximity of measured data points. We present our SILC framework in a tutorial fashion, providing the essential lifted- and frequency-domain system formulations and stability and performance criteria. A simulation-based demonstration using an empirically derived model of a microscale AM system examines SILC update law designs. Importantly, as a spatial sensor for an AM system will record as many as 108 measurements, we demonstrate that a frequency-domain framework reduces the computation time by three orders of magnitude and increases the tractable number of measurements by three orders of magnitude, in comparison with the lifted-domain framework.