Figure 1 Four levels of NDE data fusion.
Discussion
A variety of process monitoring and nondestructive evaluation (NDE) tests are applied to many complex components during manufacturing, repair, and inservice inspection, frequently with little or no cross correlation. It is not unusual that multiple NDE test methodologies will be applied in a sequential process to obtain a final decision on the disposition of a component or structure. Often each test method or mode is performed independently. This is particularly true for large aerospace components, in part, because of the distributed nature of the manufacturing processes, test facilities and compartmentalization of test and analysis skills. The relationship between one measurement mode results and another may or may not be fully appreciated. With digital computing power and automated digital test data acquisition systems becoming available enormous strides have been made in reducing ambiguity from nondestructive inspections. The availability of large data sets in digital form leads naturally to the possibility of combining and simplifying data systematically.In its most general form, data fusion is the process of integrating the data from diverse process monitoring and nondestructive observations of an object into a consistent description of the condition of the object. For example, a nondestructive test (NDT) Level II inspector performing an eddy current inspection of an aircraft skin might "fuse" his visual observation of the area of the part undergoing evaluation with a sequence of visual observations of the impedance plane display (oscilloscopic trace) into a hand drawn or "chalked-on" description of the condition of the skin. In a more complex case, data from two or more different kinds of tests (i.e., ultrasonic, shearographic, and radiographic) might be acquired for the same object by inspectors located in different states, which need to be combined by the project engineer into an engineering assessment of the condition of the part. In either case, the objective of data fusion is to coregister, collocate, and combine the data into a more useful form. By automating parts of the data fusion process, not only is it possible to reduce technical labor requirements, but also variability in the evaluation process is reduced.
Combining carefully selected "modes" or alternative views of process monitoring and NDE data often leads to reduced ambiguity, but usually increases inspection and evaluation labor unless automated data fusion tools are available. For applications where the ambiguity of a single mode test is low and the cost of misdiagnosis is not extremely high, this additional cost may not be justified. However, when single mode test data are inconclusive and/or the cost of misdiagnosis is extremely high, employing multiple process sensors and multimodal NDE methods may be the least expensive approach to reducing ambiguity to acceptable levels.
Combining data from different process steps and NDE tests requires specialized technical capabilities beyond the requirements of conventional data analysis. In particular, cross-registration techniques and techniques for normalizing, cross-calibrating, and resampling are required. Of these, systematic registration of data with respect to the evaluated part is the most demanding requirement.
The appropriate portion of the data fusion process which justifies automation depends on many operational factors, so the key is to identify the appropriate inspections and degree of data fusion required to meet program needs. These range from developing simple tools to acquire multiple process monitoring and NDE data sets on a common workstation to tools using detailed correlation models and finite element models of the structure being inspected. For simplicity, these can be divided into form levels ("degrees") of data fusion as illustrated in figure 1. In level I data are combined from two inspections onto a common workstation and interrogative tool. At the second level, simple transforms are applied, either by the user identifying common fiducials in the data, or by pre-programmed knowledge of the geometry of the two inspections, enabling rough overlay of the data for positive identification of matching indications. When more precise coregistration is sufficiently critical, or inspection data contain complex three-dimensional information, numerical models of the structure may be employed to permit precise cross-registration of NDE data. At the current state-of-the-art is the fourth level of NDE data fusion, where prior analysis has led to a multiparameter model which combines two or more measurements at a single location into a single, more meaningful image.
Conclusion
NDE and process data fusion can occur in many forms, some relatively simple and some very complex. As the complexity and performance requirements of a product increase the need for data fusion can justify the additional cost of implementation. Typically, a data fusion workstation is used. The data fusion workstation accepts information in a variety of formats, providing a platform to interpret the data in combination or simplified presentation modes. The interpretation of data is improved by knowledge obtained from both in-process monitoring and multiple inspection measurements, leading to reduced risk of failure.Practical process monitoring and NDE data fusion requirements encompass great variety. Variations occur in data typos, formats, interpretation, dimensionality, operator sophistication, and throughput requirements. Each of these parameters affects the design and/or selection of data management tools and the potential for justifiable return on investment. To be of greatest benefit, a data fusion workstation must be developed which is suitable for locating in an manufacturing, operations and maintenance facilities. The workstation hardware must be compatible with the facility requirements of the operations environment, including space available, power, and cleanliness. In designing the user interface, the workstation and associated software must suit personnel most likely to use the workstation so that the workstation is seen as an asset and not a hindrance. For routine NDE personnel use, simplicity and robustness are required. For certain engineering and/or investigative uses, access to more complex tools is required. The user interface must be intuitive to learn, providing tools that solve the real problems conveniently. Visualization tools are used to handle drudge work-estimation of position, size, etc., in part-based engineering coordinates. Often-repeated mechanical steps are automated, such as locating data, inputting and saving data, and compiling reports. Finally, the workstation must be capable of performing its functions in a timely fashion, consistent with manufacturing, operations and maintenance needs.
A data fusion workstation has been developed using INDERS [ 1,2] data processing philosophy and commercial scientific visualization tools. Data fusion of NDE and process monitoring measurements is proving to be a useful tool for advanced product evaluation.
References
1. James Nelson, David Cruikshank and Stuart Galt, "A Flexible Methodology for Analysis of Multiple Modality NDE Data," Review of Progress in Quantitative NDE, Vol 8A pp. 819-826 1989.2. Nondestructive Evaluation Integrated Data Reduction System (INDERS) for Nozzle Components Software, Boeing Document, D180-31159-1.
Acknowledgement
This work is sponsored by Wright Laboratory, NDE Branch under contract F33615-95-C5234. Laura Mann is the WL program manager.This is a work of the federal government and is not subject to copyright.