When we combine several sensors, with or without actuators, there is an advantage. More sensors can help to make the system more robust, or more sensors can be used to get more information about the subject of interest. This page describes Sensor-Actuator Systems and focuses on the topology to make robust implementations. The tricks and examples are mainly from the Perspective of Ambulant Health Applications and systems, but can be used in other fields as well.
In figure 1 this is indicated by three steps. The quantities we can measure are normally the physical or (electro)chemical quantities. With signal conditioning these quantities are cleaned-up to represent quantities with a meaning. For example, temperature, strain, acceleration, optical transmission, pressure and skin potentials are quantities that are still in the physical world.
Based on these physical quantities, we can determine physiological quantities that have a clinical meaning. For example, the skin potentials have become ECG or EMG and represent heart or muscle activity. The pulsation in the optical transmission of a bodily part can be interpreted as blood pulses (PPG), and with some wavelength variation into oxygen saturation in the blood. This is still a deterministic step and needs signal processing like envelope detection or calibration towards a meaningful physiological property.
The third step is normally no longer deterministic, but needs some sort of property estimation. An example is to correlate burned calories to accelerometry. To do this, we need knowledge about the use case because we can not calculate calories from motion. If the estimation is not accurate enough for the application of interest, it may help to include a second or third property in the estimation like body temperature, heat flux etc. However, this is the most fascinating stage of signal analysis. Who could expect you can estimate subjective properties like stress and pain?
The technological shift that is needed is conform the Data-Information-Knowledge-Wisdom (DIKW) triangle1) expressing how raw data can be combined into a higher value resource of information. We need multiple sensor sources (polygraphy) to enable estimation of behaviour, which could never be measured with a single sensor in ambulant situations. This triangle is shown in figure 2.
Sensors and actuators for ambulant health systems can be seen as information channels. In fact, the data flow can be basically in three directions depending on the application scenario as shown in figure 3.
First, textiles can monitor a person and send the information to the outside world. This is a scenario applied to working people at risk, to monitor their condition and provide reports to a central control room. It is also an important data direction for personalised healthcare systems where a patient at risk or under test is being monitored. Our healthcare system has reached a level were the type of conditions we want to observe are occurring infrequently (epileptic seizures, early uterus contractions during pregnancy, heart failures, etc.), or are associated with behavioural and psychological aspects (stress, compulsive disorders, sleep, etc.). Therefore, in this data stream direction, the person being monitored should be unaware of the presence of sensors and electronics, in order to conduct his life or occupation as normally as possible: we don’t want to affect the life of the person using or wearing the system.
In the second information flow direction, the environment is monitored and the information or response is given to a person by means of intelligent textiles. This direction of data flow opens opportunities for augmented senses: we can make people experience information content that is normally not conceivable with our human senses. In other words: data from a source that cannot be sensed because it is out of our reach (sensory-wise or location-wise) is converted into an information modality that can be sensed. Unlike the previous category currently there is not an application example available. However, it is used a lot by designers to explore the option of bringing emotions and intimacy from one person to another over larger distances. Many of us are already close to it when carrying our mobile phone in our trouser pocket. By this we can feel when the phone rings or when a message is received.
The third information stream that is observed is where an individual person is monitored, and his own bodily data is offered in another modality to the same human body. Examples of such a short loop data stream is found in (bio)feedback systems. A well established feedback system is on the market by the Polar company2). Their watches are receiving data from a textile chest belt that measures heart rate. This enables professional and leisure sportsmen to improve their training program. Heart rate is not easily sensed by our human senses. With such a feedback concept we can improve and perform better.
Note that not all product scenarios will fit into the categories represented by the three data flow directions, simply because in some cases, the electronic system does not carry meaningful information - it just actuates some energy. Examples are electronic clothes or stockings for heating the human body and textiles with built in light for curing light-affected diseases or for wound healing.
When combining sensors and actuators in systems, the most common principle to be used is feedback. In a feedback loop the result of an actuator is measured, and the sensor reading is used to evaluate the effect of the actuator. Based on the outcome, the actuator operation is adjusted. The most common example is the thermostat in our houses controlling the temperature. Based on the desired temperature setting and the measured temperature, the heating system is switched on or off. This results into a room temperature with accuracy close to the accuracy of the temperature sensor, which would not be an option with a feed-forward system without a sensor. In fact every system that should be safe and stable consists of feedback loops. The physiology in our body, like our oxygen control, our temperature, our blood pH, everything is controlled by feedback. It is therefore a good design choice to implement feedback loops in the sensor-actuator systems for protection and safety. The measurement of core body temperature by means of heat flux is an example of a feedback method3).
The feedback loop may include the user, in which case we speak of biofeedback. An example of biofeedback is found in systems for psychological stress estimation and control for relaxation4).
With sensor arrays, we can do differential measurements. The advantage of a differential measurement is that common influences are invisible: the output is only determined by differences between sensors.
When conceptualizing a differential measurement as a set-up where two sensors are measuring at the same time but at a different location, it is also possible to measure at the same location at two different events. In a Stimulus-Response measurement5) 6), a sensor reading is compared to a sensor reading of the same sensor, but after an actuator has changed one single condition in the neighbourhood of the sensor. In that case, it can be assumed that disturbing factors changing slower than the interval between the two measurements are cancelled out. Sometimes, we can learn new parameters by monitoring the dynamic time response on the applied disturbance.
The most straightforward advantage of multiple sensors with respect to safety critical systems is the redundancy. By having more statistics about a subject, the extra information results into information about the validity or can be used to detect whether one of the sensors has become unstable. However, there is a deeper advantage of multiple sensors that may lead to the estimation of new parameters. This is studied in the field of multivariate analysis and explained in the next subsection.
While the aspects of the previous section are mainly on the sensor-head level, we also have a signal processing or mathematical layer where we can apply some multi-modal sensor concepts. Dittmar demonstrated the power of multivariate analysis with a shooting experiment using ‘polygraphy’ where many physiological body parameters are measured7). Based on the combination of responses of physiological parameters, it was possible to distinguish between a successful and an unsuccessful shot. So, a parameter like ‘hit’ or ‘miss’, which can not be measured with any single sensor on the human body, can be estimated using multivariate analysis.
To make decisions based on many parameters, the mathematical toolbox is obtained from a technique called Surface Response Methodology8). With this technique, a set of strategically chosen measurements is used to explore the multi parameter space. The optimum location is subsequently chosen form the estimated model of the parameter space. This technique was successfully applied to find new parameters in complex multi-sensor systems9).
Fuzzy logic was developed by Lofti Zadeh in 1965, and has evolved to an alternative to the binary logic of the classical propositional logic. Fuzzy logic uses soft decisions, which means that together with the parameter of interest the validity of the parameter is reported10). In situations where we have a mix of decisions with different levels of importance, Fuzzy logic can be interesting.
Besides the interaction model between sensors and actuators, the topological mapping of the network should be considered. The snowboard jacket called ‘The Hub’ by O’Neill in the winter of 2004/2005 was mainly a design exercise to deal with network issues in electronic textiles. The jacket had partially integrated buttons and wires for an MP3 player, but made use of a Bluetooth connection for communicating with a mobile phone. The Nike ACG CommJacket of 2004 had a similar approach. The Levi Strauss RedWire DLX Jeans which is iPOD compatible is still on the market just like the Scottevest Revolution jacket11). However, although these multi-device apparels are all conceptually strong for carrying entertainment products, it will be a big challenge to find the right network topology for textile safety and protective systems.
The next steps in the developments of sensor-actuator systems for ambulant health applications should come from two sides: the technology push (materials and meythods) and the concept development for applications. Electronic technology is in principle low-power and small enough to fit everywhere on the body and in our natural environment. The parties nivolved have only recently found each other:
At the end of the 1970s and the beginning of the 1980s, the field of micro electromechanical systems (MEMS) originated from silicon technology. It was seen as the next logical step from integrated electronics towards surface-micromachining and bulk-micromachining integrated with control electronics. Already in an early phase, people managed to make micropumps in silicon for moving fluids. This initiated the trend towards microplumbing on wafer-scale by the creation of fluidic channels. Since then the term microsystem technology (MST) has become more common than solely MEMS.
Terminology of microsystems can be explained based on a time line of figure 16. It all started with the invention of the transistor in 1947 (Shockley, Bardeen, Brattain) and the IC in 1958 (Jack Kilby, Texas Instruments). These inventions resulted into the product line of microelectronics. In the 1970s, people started to make mechanical structures in silicon using the technology of microelectronics. Such devices were calles Micro ElectrMechanical Systems (MEMS). Soon after, the first silicon micropumps were reported and we could observe microplumbing in silicon. This has resulted into the field of MicroSystem Technology (MST). In 1990 we saw the next expansion to fully integrated systems including mixers, chemical sensors and on-wafer separation columns. This was called Lab on a chip or µTAS (Andreas Manz, Imperial College, London). Nowadays, we don’t bother so much on which physical or chemical domain is in the system, the problem is how to package it. This approach is called System in package. A defenition of a SiP would include that it is a consumer ready which means that the combination of microelectronics with mechanical or chemical interfaces is offered to the customer. SiP’s are multi-domain and cheap due to batch processing12).
The result is that devices like microphones, accelerometers, biochemical reaction chambers, fluid pumps, fluid mixers, etc. have become extremely small, cheap and fast. This has its impact on our society where sensor and analysis systems are everywhere. In principle, all of our body information is available in the gases, fluids and electromagnetic waves on the surface of our body. MEMS technology can analyse this body information and use it to assist in making decisions to lower the risk of becoming ill or injured. With nanomaterials the level of integration can be even a step further and more sophisticated. So the trend of integration and miniaturization does not stop at the micron level. The integration level will be higher and higher until technology is completely integrated in our body and daily environment.
The shift towards wireless technologies makes it possible to connect easily to other objects. This cloud concept will transfer information from human to human, but also from any human to any physical object. The ‘Internet of Things’ is coming13), and the impact of having all Internet data available at any low-profile node will change our life. What once started with tags in anything will expand to a world where the position and state of any object is known. It will not only be known for a scanner close to the object, but also to any other object wherever in the world. Data is not anymore stored at localised computers but data will be decentralised and distributed.
Just like we could estimate new immeasurable parameters from polygraphic multi-sensor networks, the global internet of objects and data will give opportunities we can not yet even imagine. Especially for protection and safety this must make a big difference. Until now, epidemiological data on diseases could only be evaluated on demand, after people got ill and after the disease was identified. With a cloud of objects and information about many people, we can not only identify diseases in an earlier phase, we can also immediately correlate the patterns to all other events in our society.
The current trend towards monitoring mental conditions will expand in future to concepts where mental conditions of multiple people can be mapped. This will be used for risk analysis because stress correlates immediately to accidents, and so, accident prediction becomes feasible.