Implementations of a Four-Level Mechanical Architecture for Fault-Tolerant Robots, Page 2

Level I - Existing drive systems usually include motor, encoder (or resolver), brake, drive train and joint bearings. They have no fault tolerance capability and could suffer failure through a variety of modes. To provide fault tolerance, one option suggested at NASA-JSC is to use multiple motors driving separate primary gears in a common gear box. This requires that all motors be capable of rotation at all times and any degradation in a motor's performance affects the joint. Another option suggested by Wu et al. uses dual motors driving a common axis through a bevel differential drive. This work suggests that such a drive system should have a high forward efficiency and low backdrive efficiency. Though this design can tolerate a failure of either motor, a failure in the differential will cripple the joint. Without complete functional duality, it is impossible to mask all the probable failure modes. It is, therefore, imperative that a dual actuator set driving a single joint be designed. The torque-summing actuator is symmetrical about the center line and has complete duality between the right and left halves. Each half contains an armature, rare earth magnets, resolvers, brakes and Ferguson's paradox epicyclic gear trains. The Ferguson's paradox gear train was chosen primarily for its high gear ratio and compactness. The module has attributes of low-weight, high-stiffness, minimal interfaces to the system controller, and overall compactness. The prototype of the module incorporates mainly off-the-shelf components compactly laid out in a configuration that provides low weight and high drive stiffness. The dual sided module provides fault tolerance in the event of failure of one side by "doubling up" on the other side by peaking its power for a short period of time. In the event of seizure, the failed side has to be detached from the power train to allow the joint to turn. For the prototype actuator, such a detachment is necessary even in the case when the failed motor is free to move because of the non-backdrivability of Ferguson's paradox gear trains. Conceptually, a dual fault tolerant actuator can be incorporated into fault tolerance at levels 11, III, and IV. The prototype module, however, is designed for testing and development as a stand-alone unit and is mounted to a heavy steel test stand.

Level II - Level II of the fault-tolerant architecture constitutes parallel structured modules, of which the Stewart platform is a well-known example. The addition of one leg to the familiar three-legged design gives this Stewart platform a redundancy of two. Knuckle structures with 2 DOF form the base joints for each leg. The knuckle is designed to tolerate a minimum of one fault before failing. The system controller acts as a supervisor in analyzing the sensory feedback with a Fault Detection and Isolation (FDI) algorithm to determine if a fault has occurred in the knuckle. If a fault occurs at the actuator level, the motor is removed by disengaging a clutch. Each servo system consists of a clutch, a brushless resolver, a brake, Hall-effect sensors, and a three-phase Brushless DC motor. Each motor has a peak torque rating of 40 Newton*meters and a continuous rating of 6 Newton*meters. There is a .5 meter spacing between the output shafts of the motors, measured along the common axis of rotation. The entire knuckle system weighs about 440 Newtons. Each motor is controlled by a Digital Integrated Servo Controller (DISC). The system controller is a personal computer operating under the Lynx O/S real-time operating system. FDI algorithm supervises the servo systems. The topic of FDI enjoys a rich history in the literature. Roughly twenty years ago Willsky' generated a survey paper presenting a number of statistical techniques for fault detection in dynamic systems. Over ten years ago Isenmann developed another survey paper on fault detection based on modeling and estimation methods. NASA also has shown considerable work in the area of FDI, including applications to a multisensor navigation system' and to the space shuttle's main engines As one would hope, FDI is also important within the nuclear power industry. Singer et al. discuss the use of sequential statistical techniques in the analysis of the primary coolant pumping system in the EBR-H nuclear reactor. An expert system using pattern recognition and fuzzy inference techniques analyzes the statistical information to provide the fault detection. More recent work includes that of Visinski, et al. on fault detection thresholds based on dynamic system models. By including multiple redundant sensors in each servo system, as well as complete redundant servo systems, the knuckle was designed specifically for implementing and testing these types of algorithms.

In the current algorithm for the knuckle testbed, the servo control systems are assumed to be fully observable, deterministic, and linear time invariant. When an observable, accurate model exists, the FDI implementation uses parameter identification methods for detecting and, in many cases, isolating faults. The FDI algorithm also makes use of the currents in the DC servo system to determine whether enough torque can be generated to provide the motion desired. Threshold levels are set to check for deviations outside of the nominal values (which are determined experimentally) and any such deviations are classified as faults. The algorithm also filters "false alarms" generated by transient responses or noise levels outside the allotted error bands. The DISC is a very compact brushless DC servo controller that offers numerous features not contained in any single commercial system available today. Some of the features include: multiple sensor interfacing, compact 'smart' power electronics, fault tolerance, and high speed digital communications designed in a modular package. The features incorporated into each DISC allows for research into the use of distributed control applied to a modular robot. Distributed control can significantly reduce the amount of wires running through a structure since the DISCs digitally multiplex the signals locally. This reduction in wires is essential to the success of modular robotics and mechanical fault tolerance. The DISC contains four sub-modules: a sensor interface module, a communications module, a power interface module and a digital controller. Sensor Interface Module: The sensor module conditions and digitizes all external signals before passing them to the controller module. The module will read multiple sensors, including: a tachometer (16 bit resolution), three current sensors (12 bit resolution), a resolver (16 bit resolution), a torque sensor (12 bit resolution), an incremental encoder (16 bit resolution), two temperature sensors (8 bit resolution), a logic level Hall-effect sensor, an auxiliary +/- 5V signal (8 bit resolution) and an auxiliary +/- 10V signal (8 bit resolution). Communications Module: The communications module controls the information between the DISCs and the system controller. The difficulties inherent in electrical interface design for modular robots are primarily a result of the high number of wires that run throughout the robot. Most robots are controlled by monolithic system controllers. All data is brought to the system controller using separate conductors. As a result, every signal, sensor, and power wire for each actuator must be run throughout the entire robot to the system controller. For example, a Cincinnati Milacron Inc. T3-776 industrial robot requires 24 wires for the tachometers, 54 wires for the resolvers, 18 wires for the brakes, 40 wires for powering the motors and cooling system, and 15 to determine whether enough torque can be generated to provide the motion desired. Threshold levels are set to check for deviations outside of the nominal values (which are determined experimentally) and any such deviations are classified as faults. The algorithm also filters "false alarms" generated by transient responses or noise levels outside the allotted error bands.  Next Page ->