The gaseous hydrogen flow control vale (GH2FCV; see Figure 1) is a critical space shuttle component that controls the pressure of the hydrogen external tank during launch and ascent to orbit. The valve is one of the components that are to be monitored by the Integrated Vehicle Health Monitoring (IVHM) system. The GH2FCV is a two-position valve that is actuated by electromagnetic force acting against a return spring. In the de-energized position, the gaseous hydrogen flow is high; when the valve is energized, flow is reduced. There are three such valves on the shuttle, one connected to each of the three main engines and manifolded together to return hydrogen gas to the external tank (see Figure 2).

Figure 1. Gaseous hydrogen flow control valve (GH2FCV).

The valves are precision manufactured but, in spite of this, each exhibits slightly different operating characteristics from the others. Each valve is routinely tested prior to flight and upon return, but cannot at present be tested during flight.

Figure 2. Location of GH2FCV.

Testing is performed by actuating the valve, recording the data, and having experts evaluate the current signature. The latter is the current-vs.-time waveform exhibited by the valve as it is energized and de-energized (Figure 3). The characteristic waveform is similar for all GH2FCVs, but small tolerance changes in manufacture and other variables make each valve unique and the corresponding waveforms different. The exact curve signature is a function of electrical and mechanical characteristics of the valve as well as many other variables. It is precisely these subtle changes in the curve signature that yield information about the status of the valve. Thus knowledge of the specific valve and its history are needed to make a good assessment.

Figure 3. Typical current signature.

This fact makes it necessary to have experts interpret and evaluate the waveforms. If any anomalies are detected, or even suspected, the valve is replaced for safety reasons. Testing, therefore, is subjective, expensive, and time-consuming. Automating this procedure, especially having the ability to monitor the valve in flight, would reduce maintenance time, extend the operating life of the valve, and improve knowledge of the valve's overall health.

To address this problem, Oklahoma State University utilized neural networks (NN) trained to recognize the unique characteristics of the individual valve and its fault conditions. Once properly trained, the NN automatically evaluates and reports on the health of the valve each time it is operated. A prototype system was designed and built by Oklahoma State University and demonstrated to NASA personnel. The prototype consists of a personal computer (PC), a smart box, and a current sensor (see Figure 4). The current sensor signal is amplified and digitized. The data is collected, with 8-bit resolution at 10,000 samples per second. When the valve is energized or de-energized, the waveform is analyzed and the results of the particular activation are displayed on a 2-line by 16-character display that is an integral part of the smart box. The PC is used only for user interface and initial NN design and training. Once training is complete, the smart box operates standalone and may be disconnected from the PC.

Figure 4. System block diagram.

To operate the system the user is prompted by a graphical user interface(GUI) to select a suitable NN structure. The structures may have multiple inputs, outputs, and layers, and any number of neurons. The smart box is switched to "training" mode, and training is done on the valve in question. Training is achieved by activating the valve and collecting current data versus time. With each activation, the data is sent to the PC, where the user sees a plot of the current signature. The user is prompted to accept or reject this plot. If it is accepted, the user then provides the corresponding target values. All desired fault conditions must also be applied. These may be real or simulated. The raw data is processed to extract critical features. Once the desired amount of training data has been collected and processed, the results are used for defining the weights and biases of the NN. This is accomplished with Matlab from The MathWorks and the Neural Networks Toolbox (both commercially available software packages can be obtained from MathWorks).

Once an acceptable network has been defined, the structure with weights and bias information is sent to the smart box. At this point the box may be disconnected from the PC and switched to the "monitor" mode. In this mode the box automatically detects activation of the valve, collects data, preprocesses it, feeds this information to the NN, and displays the results.

The prototype system was designed for laboratory testing, but with the installation of a passive current sensor on the valve, it can adapted for flight as well. The flight data could then be used by the IVHM system for better understanding of the health history of the GH2FCV and to help predict future failures. The types of problems that the system can learn to detect include sluggish poppet, supply voltage problems, coil resistance variances, etc. Although designed specifically for the GH2FCV, the concept can be used to test similar valves that are used in the space shuttle and in other non-space critical systems. Having real-time information regarding the health of the GH2FCV reduces maintenance time and cost and extends the usable life of the valve.

This system was designed by Dr. Carl D. Latino, Oklahoma State University, for NASA Kennedy Space Center's Advanced Developments and Shuttle Upgrades. The hardware was constructed and software written by students at the OSU Systems Prototyping Laboratory. The system was installed, refined, and demonstrated at NASA KSC with the help of Mario Bassignani and other KSC employees. For more information please contact Dr. Latino at (405) 744-5151; fax: (405) 744-9198; e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it..