Multi-dimensional and tomographic measurement

Citation

Hoyle, B.S. (2000), "Multi-dimensional and tomographic measurement", Sensor Review, Vol. 20 No. 3. https://doi.org/10.1108/sr.2000.08720caa.002

Multi-dimensional and tomographic measurement

Multi-dimensional and tomographic measurement

The notion of a sensor typically conjures up in our minds a sophisticated engineering product, that at heart exploits some physical property in order to transform a process variable into an electrical value. Such sensors are ubiquitous in modern industrial processes - they deliver the key information that enables its successful management and control. Sensor manufacturers have innovated impressively to provide auto-calibrating, linearised and self-testing sensors, so that the elementary physical transduction at the core of the sensor is now embodied to form a precise single-point measurement component.

In more considered terms such a sensor must still sample its target variable in space and time. Like all sampling operations, this is only effective if it is representative of the underlying process. For many processes this is reasonably simple to arrange in spatial terms. The pressure in a vessel holding a liquid, a specific dimension of a moulding, or the temperature of a surface, can comply with this condition in spatial terms. Obviously sensors can be placed at carefully selected points to ensure representative information. Similarly, most simple measurements can be made often enough to estimate any dynamic changes in the process.

However, where the process has complex features or many aspects, it is more difficult, or even impossible, to devise a representative sampling operation from a single point sensor. It may be tempting to assume that we can make an approximation to a representative sample in many cases; however, in doing so we inherently accept limitations in the operation of the process. Thus we will constrain the process to operate only within a specific narrow envelope, in which the sensors can be assumed to provide an outline indication of the surface, or internal state. When operating conditions are changed, perhaps away from initial norms, this single point approximation is likely to be reduced in value, or may even be misleading.

In order to break out of this limited view we must first examine the opportunities liberated if multi-dimensional sensors were available. It is clearly valuable to gain the deeper understanding that a more representative multi-dimensional knowledge of a process or product can provide. Industry in general has many facets and myriad needs for measurement. We can attempt to classify this need in many ways. A simple classification may be based upon the type of physical measurement while lies at the heart of the measurement process.

In this edition of Sensor Review we choose a classification that looks uniquely to the future of measurement: where this embraces several dimensions sensed in a single operation. This viewpoint is becoming increasingly viable, due to the combination of a range of new sensing technologies with the processing capability to transform the complex sensed data into meaningful information within the requisite time-frame.

For example in the case of the physical measurement of length, industrial challenges are typically concerned with complex metrology problems. Where a simultaneous multi-dimensional measurement is possible, this may offer large benefits in verifying shape; even to the point of enabling new processes in which a forming process is controlled by the sensors, in contrast to a post fabrication go/no-go check. High accuracy representative data in such cases may be obtained through the precision with which we can detect optical and laser beams.

In recent years we have seen a large increase in the use of vision-based systems, in which essentially 2D data is sensed and processed, typically in close association with an expected model for the manufacturing process. Systems that deliver 3D information are now emerging to extend this trend. Computation of 2D information is typically obtained by optical scanning followed by complex model fitting. For example we can detect the alignment of a part to be placed, or measure a dimension through intrinsic image calibration.

A 3D inference demands a more complex technique, either through the use of stereo vision, (as embedded in a human being), or through other more demanding techniques such as multiple time of flight measurements. In both cases fast processing is essential to provide a measurement within the requisite dynamic time-frame. We can expect major developments in these techniques. Innovative manufacturers will harness them to develop new ways to fabricate their products to higher specifications with reduced waste and costs.

Where products are manufactured in bulk form, such as a liquid or powder, we find a need to characterise an internal space, typically occupied by a mixture of materials or reagents. This is of course also the case when we examine the manufacturing processes for salient products. In most cases these comprise a range of complex states and phases. For example, industrial mixers, separators, multi-component pipelines, filter-beds, strip processes, extruders, polymerisers, and die injection systems for plastic moulding.

A primary key to efficient design and flexible operation of such industrial manufacturing processes is high quality information concerning their actual internal state. Many process systems are incompletely understood; are designed for operation under specific narrow conditions for continuous operation, or for use only with particular types and grades of raw materials. Such information gaps preclude more flexible use of the process to meet new markets, or tighter environmental standards. Industry is under increasing pressure to utilise resources more efficiently, and to satisfy demands for product quality and reduced emissions.

The internal state of a typical process can be viewed as a 2-dimensional cross-sectional distribution of the key parameter of interest, or, for large process vessels, in terms of the 3-dimensional distribution. When the need for process information justifies costs, multiple sensors could provide the requisite 2D, or 3D information. The estimation of the internal distribution of key parameters in industrial processes, from such measurements, is in essence that of process tomography (PT).

The background science of PT has evolved from medical tomography, enabled by modern data processing systems. In essence, a sensor array is used to obtain cross-sectional tomographic data, which are reconstructed to form a pseudo-image of the internal state of the process. The image is then interpreted to provide process information: component volume fraction in flowing mixtures; solids concentration in stirred reactors; density distribution in a product, etc. These operations are repeated to reveal continuously the behaviour of the process of interest, usually at a rate appropriate to its real-time dynamics. Industrial applications of tomography are more demanding than their medical antecedents: typical processes are much less homogeneous than animal tissue, necessitating intensive reconstruction processing.

Such tomographic sensor systems enable the distribution and movement of fluids and multi-phase mixtures in process vessels and pipes to be sensed in two and three dimensions, thus offering a new and radically different approach of potential benefit across a wide range of manufacturing processes. Unlike medical tomographic imaging, the sensor methodologies must be low cost, robust and able to sense rapid dynamic changes within the processes. A number of systems have been developed using a range of basic sensing techniques. For example electrical capacitance may be used in non-conducting processes, electrical resistance in aqueous mixtures, ultrasound in multi-phase mixtures, etc.

Process images, computed from sensor data at real-time rates appropriate to the process dynamics, are typically analysed automatically to derive appropriate information on the process dynamics for monitoring and control purposes. Although such images are available for viewing, perhaps for verification of CFD models, they are also valuable as the raw data for a following interpretation process. For example a stirred reactor may be characterised in terms of a set of mixing ratios. A particulate process may be similarly characterised through homogeneity indices. As in the case of 3D external image sensing, the internal sensing offered by process tomography will enable innovative approaches to process design.

Brian S. HoyleGuest Academic Editor