Subramaniam Balakrishnan

Sequencing jobs on a single machine: A neural network approach

Abstract This paper presents an approach for single machine job sequencing problems that is based on artificial neural networks. A problem is classified first by one type of neural network into one of a number of categories. The categorization is based on the problem’s characteristics. Then another neural network, which is specialized for a particular category, applies a previously ‘learnt’ relationship to produce a job sequence that aims to better satisfy the given objective. The learning is acquired in these networks after a training process in which the network is exposed repeatedly to a set of example problems and their solutions. The trained network thereby learns predominant relationships between given problems, and the output sequences that optimally meet the desired objective. The advantage of such an approach is that it allows what amounts to a ‘customized’ heuristic to be established for problem subsets and various objectives without having to deduce an algorithm in advance. The methodology and its implementation is described for several of the more common sequencing objectives, as well as for a hypothetical objective that minimizes a cost function exhibiting a limited exponential behavior.

A generic deadlock-free reactive scheduling approach

In this paper, a generic deadlock-free reactive scheduling (GDRS) approach is proposed. The approach can react to the occurrence of a variety of disruptions in flexible job shops, in real time. Internal and external disruptions such as machine breakdown, process time variation, urgent job, arrival of a new job order, and order cancellation are considered. Performance comparisons of the proposed approach with that of Total Rescheduling (TR), and with the modified Affected Operations Rescheduling (mAOR) heuristic proposed earlier in the literature are presented by considering scheduling efficiency and stability. The results show that, for most of the disruptions considered, GDRS retains the rescheduling efficiency of TR while providing superior system stability over the other two approaches.

A Search-Based Heuristic For The Two-Dimensional Bin-Packing Problem

AbstractA heuristic algorithm combining priority rules with a restricted search procedure is presented for solving the two-dimensional, bin-packing problem. The method of solution is one of problem reduction, where the initial problem is decomposed into subproblems that are solved separately. The search procedure acts to minimize the number of subproblems created in the decomposition, while the priority rules force the search process to give due weight to the less combinatorially-inclined pieces. The algorithm was tested using four heuristic rules for assigning priorities, and data from two previously published algorithms. A good improvement was generally observed.

Deadlock-free scheduling of flexible job shops with limited capacity buffers

In this paper, Mixed Integer Programming (MIP) formulations of the deadlock-free job shop scheduling problem are proposed. The presence of buffer space with limited capacity is considered. This research work also proposes a novel operations insertion algorithm based on the rank matrix (or Latin rectangle). In this algorithm, rank matrices are used to generate the schedules and to check for deadlock situations. Finally, an insertion algorithm is proposed to insert transportation operations in the obtained schedules. Performance evaluations of the proposed mathematical models and the proposed algorithm are conducted. The results show that the mathematical models outperform a model presented earlier in the literature. The results also show that the proposed algorithm obtains the same or better solutions when compared to other solution methodologies reported in the literature. [Submitted 31 July 2007; Revised 14 October 2007; Accepted 14 October 2007]

Mathematical formulations for scheduling in manufacturing cells with limited capacity buffers

In the past decade, the deadlock-free scheduling problem in flexible manufacturing systems has received much attention from researchers and practitioners. This is due to the growing trend of automation, and the rising need for flexible manufacturing systems that can cope with the everyday changing market demand. In this article, mixed-integer programming formulations for the deadlock-free scheduling problem of flexible manufacturing cells are proposed. A job shop environment is assumed where each job may have a different processing route. The proposed models consider the presence of different types of buffers in the system. Furthermore, to enhance the comprehensiveness of the models, a heuristic to insert transportation operations into the obtained schedules is proposed. Finally, computational experiments are conducted to investigate the performance of the proposed models in terms of efficiency and computational time.

Haptic-enabled teleoperation of base-excited hydraulic manipulators applied to live-line maintenance

This paper investigates haptic-enabled teleoperation of a base-excited hydraulic manipulator working under a wireless communication channel. The intended application is live power line maintenance. With respect to this application, three main challenges are recognized in the field: need for the force feedback, wireless communication between master and slave sites, and base excitation of the slave manipulator. In this paper, a test rig is developed to examine how an operators hand speed regulating scheme enhances the linemans performance while the entire system works under a wireless communication channel and the slave base is under excitation. Two sets of experiments are performed when the haptic device produces no force and when the regulating haptic force is added to the master device. Performance of the system is evaluated by measuring four indices: operators failure in completing a task, end-effector displacement, slave manipulator controller effort, and task completion time. Results indicate that adding the haptic force to the system helps linemen function more effectively when the system is subject to base-excitation and communicates through a wireless network.

Surface roughness measurements in finish turning

A new approach is proposed for the on-line measurement of the maximum peak-to-valley roughness,Rmax, of a finished-turned surface in the feed direction. The method is based on solving the inverse problem of light scattering by using a linear least-square estimate of the angular scattered light pattern reflected from a surface. A laser system has been developed to capture the light reflected under different cutting conditions. The effects of the ambient room light as well as the workpieces rotational speed and methods for thier compensation are also discussed. Good correlation was found between the optical and stylus-measuredRmax.

Application of neural networks for surface roughness measurement in finish turning

A laser system which incorporates a charge-coupled-device (CCD) sensor is developed to measure, in real lime, the maximum peak-to-valley surface roughness, R max. produced during finish turning. A three-layer neural network is used in conjunction with a back-propagation learning algorithm to predict R max, by quickly recognizing the angular scattered light patterns (ASLP) reflected from the workpiece in the feed direction. The predicted R max values have a maximum error of about 10% when compared to conventional stylus measurements.

Intelligent robotic assembly

Abstract It is normal when programming a robotic manipulator to provide the end effectors orientation and position at the pick up and drop off locations. Additional sensory information and intelligence is needed, however, to detect the presence of a part as well as its location if the assembly site cannot be controlled precisely by employing expensive jigs or fixtures. This paper investigates, for this purpose, the application of solely an inexpensive laser sensor mounted unobtrusively to the end effector of a CRS robot having customized hardware and open software. Data from the sensor is converted into a single “Feature Value Vector” to recognize a part and accurately determine its location by using a neural network and back propagation training. The procedures viability is tested by assembling a set of tightly meshing gears under poor ambient lighting.

A prototype telerobotic platform for live transmission line maintenance: Review of design and development.

This paper reports technical design of a novel experimental test facility, using haptic-enabled teleoperation of robotic manipulators, for live transmission line maintenance. The goal is to study and develop appropriate techniques in repair overhead power transmission lines by allowing linemen to wirelessly guide a remote manipulator, installed on a crane bucket, to execute dexterous maintenance tasks, such as twisting a tie wire around a cable. Challenges and solutions for developing such a system are outlined. The test facility consists of a PHANToM Desktop haptic device (master site), an industrial hydraulic manipulator (slave site) mounted atop a Stewart platform, and a wireless communication channel connecting the master and slave sites. The teleoperated system is tested under different force feedback schemes, while the base is excited and the communication channel is delayed and/or lossy to emulate realistic network behaviors. The force feedback schemes are: virtual fixture, augmentation force and augmented virtual fixture. Performance of each scheme is evaluated under three measures: task completion time, number of failed trials and displacement of the slave manipulator end-effector. The developed test rig has been shown to be successful in performing haptic-enabled teleoperation for live-line maintenance in a laboratory setting. The authors aim at establishing a benchmark test facility for objective evaluation of ideas and concepts in the teleoperation of live-line maintenance tasks.