Chapter 1 - Introduction

Back to Abstract.

Faults within large scale systems containing servo systems are often more expensive than the servo systems themselves. Therefore, a servo system fault can cause a very costly or catastrophic event. To prevent such an event, on-line and off-line diagnostics are often used.

One example of this type of servo is the yaw-damper servo-actuator used in the C-141 aircraft. This servo is flight critical. Therefore, when it is not functioning, the whole aircraft is grounded until it is fixed.

The servo systems have been popular for a long time due to their ruggedness, simplicity, and diverse applications. However, servo systems do fail and wear out. Therefore, the research of this thesis is based on the merger of two basic ideas:

1) The reduction of aircraft maintenance cost

2) The improvement flight-critical system safety.

The result is the use of predictive diagnostic procedures to lower maintenance cost while improving flight safety in a flight control application of a servo-system. Example applications include the C-130, C-141, C-17, and V-22 servo systems which use electrical servo-mechanisms. Applications are possible for servos with and without hydraulics.

1.1 Predictive Diagnostics Design Goal

The pilot and maintenance teams must be considered to formulate the goals of diagnostics and prediction.

To satisfy pilots, the system must be reliable and understandable. Thus the system must detect or predict all failures which adversely effect the aircraft performance.

To satisfy maintenance teams, the system must provide the same fault detection, and isolation to the Line Replaceable Unit (LRU). Since almost all of the servo-systems on an aircraft are contained within LRUs, the LRU isolation is readily achievable through standard computer hardware design techniques. Lower levels of isolation are not necessary for on-board application because the repair of failed hardware is normally performed at a central depot.

To detect or predict failures which effect system performance, the servo performance parameters must be measured, then extrapolated. Figure 1 shows the block diagram of the fault prediction system.

Figure 1, Fault Prediction System

 Although it is possible to use a high order system model in the Least Squares Minimization (LSM), a second order approximation is best. Use of a higher order model would require more CPU time from the Flight Control Computer (FCC). Use of a first order model may not be adequate for some servo systems. The second order tendency principal17 is used to justify the application of the second order model to higher order systems.

1.2 Maintenance Costs

In many applications, maintenance cost is considerably higher than the procurement cost of a servo system. For example, a failed C-141 yaw-damper servo prevents deployment of the aircraft until an unscheduled maintenance action can repair the flight-critical servo. Thus, the aircraft owner periodically suffers the costs of down time, a maintenance team and replacement parts.


In the example case of the C-141 yaw-damper servo, the cost of down time is hundreds of times more than the cost of the replacement parts. The cost of the maintenance team is also much more expensive than a typical servo system.

Historically, the U.S. Air Force has purchased detailed maintenance procedures written in the form of Technical Orders (TOs). The TOs contain detailed instructions for trouble shooting. Many of the trouble shooting procedures are precisely made up of the steps:

1) removal

2) disassembly

3) inspection

4) reassembly

5) reinstallation.

These TOs are still the most common method of maintenance for the Air Force.

The most notable efforts to reduce the maintenance cost are focused on replacing older components of the systems with more reliable newer ones. This approach is reducing the maintenance cost, but there is a limit to the effectiveness of this approach.

The goal of this thesis associated with maintenance cost is to predict wear-out type failures of servo systems controlled by computer systems onboard the aircraft. With the predicted fault detection, scheduled maintenance can then be used to prevent down-time in the future.

1.3 System Safety

Another related field of research into flight control systems is the area system safety. In the flight control application, fly by wire applications are spurring efforts to increase flight safety, through fault tolerance, redundancy, and diagnostics.

The most common approach to achieve a good system safety level is through redundancy. In flight critical systems, redundancy management techniques are used to reduce the probability of aircraft loss below 1/109. Typical systems include the quad-redundant flight controls of the B-2, F/A-18, V-22, and C-17A.

Obviously the cost of redundant systems on aircraft is very high. For example, the cost of a quad-redundant system is more than 4 times that of a simplex system. First the system is multiplied by four, but then additional circuitry and software must be added to handle the communications between the computers of a redundant system.

The goal of this thesis associated with system safety is to predict wear-out type failures of servo systems in flight which are controlled by computer systems onboard the aircraft. With the predicted fault detection, the pilot can take action (such as landing) to prevent catastrophic events. With system safety increase, the needed level of redundancy is reduced. In some applications this could reduce a quad system to a triplex, or a triplex system to a dual, where a significant cost reduction would be achieved.

1.4 Application of Predictive Diagnostics

Methods for servo-system diagnostics are very diverse. However, the field of predictive diagnostics is relatively new.

Predictive diagnostics are based on the determination of the system's physical condition, and projecting trends against a set threshold.

Classical methods for determination of system physical conditions are based on disruptive test procedures. These are commonly performed as either go/no-go tests, or maintenance tests. For example, the "Blocked Rotor" test is widely used to determine the characteristics of the armature of a motor. This test is often performed as a last check before removal to insure the system failure has been isolated properly before a component is replaced.

The next level of diagnostics is system monitoring. System monitoring is based on automatic testing. The purpose of system monitoring tests has branched out into the two areas:

1) maintenance tests

2) confidence tests.

Maintenance tests are used to provide fault detection and isolation information to maintenance operators. On-line information reduces the fault detection and isolation times for system maintenance.

Confidence tests are intended to give system operators (pilots) confidence that the system is working properly. The first confidence test to gain wide popularity in flight control was the pre-flight test. Some of the modern on-line confidence tests check the system on a noninterference, periodic basis, checking for static failures.

System monitoring tests which predict the performance level have not yet been adequately developed for deployment.

To apply predictive diagnostics, parameter estimation and trend analysis are necessary. The parameter estimation provides a time series of performance measurements. The trend analysis is used to project the current and past performance into the future.  

1.5 Related Research

The first area of related research is in the application of predictive techniques to the feedback compensation.

In Predictive Compensation Of Visual System Time Delays by D.J. Sobiski1, the predictive algorithm was used to compensate for the system time delay parameters of a digital flight control for maximum aircraft performance. This research is relevant because of it's application of a predictive algorithm to a simulated aircraft attitude with a pilot in the loop, and varying aircraft parameters such as load. This research employs predictive parameter estimation into the control loop. The parameters estimated for the plant are used to modify the control loop.

A sub-field within the area of feedback compensation is to develop fault tolerant systems.

In Reconfigurable Flight Controller for the STOL F-15 With Sensor/Actuator Failures by D.L. Pogoda, and P.S. Maybeck2, techniques to maintain control of an aircraft without an operating servo-actuator are developed.  

The second area of related research is in the application of diagnostic techniques to achieve fault tolerance.

In Computer-Based Diagnostic System For A DC Generator by R.W. Devine3, estimation algorithms were used to isolate faults within a DC generator system. The model parameter estimation was successful, but the isolation seems questionable. His approach is to use a pattern recognition system to determine what has failed. Fault patterns were identified, but not fully demonstrated.

The isolation problem was due to the fact that many failures can have the same effect on system operation and thus isolation to the level attempted may require a large amount of CPU time to be spent with fault pattern analysis. There is also a fundamental flaw in the criteria used for a fault. The criteria of a fault was any mechanical or electrical parameter which changed by more than 10% from it's nominal value. The problem with this assumption is that not all 10% variations matter. For example, a 10% shift in brush resistance has no noticeable effect on the system performance. If the system were to annunciate a fault because of 10% variation in brush resistance, the pilot would loose confidence in the whole diagnostic system because nothing would feel like it was failed.

In Efficient Fault Isolation Schemes for Grey Digital Systems by M.H. Hede4, intermittent and digital fault isolation is developed. Since a predictive system may have to deal with intermittent faults, they must be considered for the controller design. The dissertation defined the term Grey Digital System as "a system in which the circuit implementation of some, or all, subsystems is unknown, but whose interconnections among all its subsystems are known." The author goes on to develop an efficient probabilistic isolation strategy for intermittent and permanent faults while ignoring degradation wear-out type faults.

In An Integrated Approach to Controls and Diagnostics: The 4-Parameter Controller by C.N. Nett, C.A. Jacobson, and A. T. Miller5, the Hilbert technique was applied to fault isolation by comparing the fault conditions against a map to determine the most likely cause.

In Fault Detection in Linear Discrete Dynamic Systems by a Pattern Recognition of a Generalized Likelihood Ratio by S. Tanaka, and P.C. Muller6, the detection of an abrupt change in the parameters of a linear system was considered. A robust detection method based on a generalized likelihood ratio was proposed. The technique is a step in the right direction, but the transition to prediction of faults was not made.

In Control of an Active Suspension System Subject to Random Component Failures by C.D. Benito7, Kalman filters were used in fault detection within a fault tolerant system. However, the filters were not predictive. State estimates were used to compute the certainty and type of the suspected fault. The importance of this research is that on-line estimation and tests were performed on servo actuators during operation.

In On-Board Real-Time Failure Detection and Diagnosis of Automotive Systems, by C.D. Benito, Kalman filters were used in fault detection without using the predictive capability of such an algorithm.


Chapter 2. Methods and models of failure prediction

Abstract - Fault Prediction With Regression Models