SIEMENS

You might say that Dr. Rudolf Sollacher is abolishing hierarchies. Sollacher, who works at Siemens Corporate Technology (CT), studies ways in which intelligent sensors can organize themselves in a network. Such networks could be used in a building to collect data on temperature, gas levels, or smoke. They would not simply pass data on to a control center, but instead jointly evaluate the situation. In other words, they could determine whether a fire has broken out in the kitchen or a pot is boiling over. Such a system would be very robust, says Sollacher, “If a central control unit fails in a fire, the information dries up. But if intelligence is distributed, various sensors will continue sending data that can be used by the fire department.”

Sensors are the eyes and ears of any intelligent control system, whether for automated manufacturing or monitoring major industrial facilities. Their data is used by control systems to make decisions and issue commands to actuators such as controllers and motors. They form part of intelligent building blocks — so-called sensor nodes — that contain microprocessors and sometimes communication or positioning devices. Thousands of these digital monitors can now be found in modern buildings. But as their numbers increase, so too does the complexity involved in programming associated controls. After all, every sensor node has to be initialized and every function — for example, the generation of a diagnostic report — must be programmed.

It would be much simpler and cheaper to have a sensor network that does all this itself. With this in mind, Sollacher’s team conducted a study that showed how sensor networks can simplify the search for building materials at construction sites covering as much as several square kilometers. Most such materials of value — things like cable spools and motors — come with radio frequency identification labels (RFID tags) that generate data that is stored in a database. But containers are moved around, machines are adjusted — and at some point databases are no longer accurate.

CT’s concept uses sensor nodes consisting of a positioning device and a communication unit. The nodes are placed on poles distributed at 50-meter intervals across a site. The nodes autonomously take distance measurements to their neighbors so that each node knows its position. They then network themselves via radio and register the RFID tags in their area. When a worker enters the ID number of the material he or she is looking for into an RFID reader, the query is sent to all of the nodes in the network. The target RFID unit then displays arrows that point the worker to its location. If the construction site expands, you simply install additional poles and nodes.

Most sensor nodes today are linked to control centers via cables. This is expensive and complicated, especially when many devices have to be hooked up, or when the units are mobile, as is the case with robots in the automotive industry. To get around this, researchers are developing radio sensors. But to be practical, such devices will have to have their own energy supply while offering secure and stable radio communication. Under Sollacher’s direction, CT is developing technologies for such radio sensor networks as part of a project known as ZESAN, which is sponsored by the German Ministry of Education and Research (BMBF).

Harvesting Energy. Sollacher, a committed microwatt miser, wants his sensors to be as energy stingy as possible. His colleague Daniel Evers is therefore studying self-powered radio sensors that literally harvest energy from their surroundings (see Pictures of the Future, Fall 2009, Tapping Ambient Energy). Evers adorns his sensors with solar cells, for example, or with piezoelectric converters that transform mechanical pressure into voltage. He also equips them with thermo- electric materials that generate energy from temperature fluctuations. A sensor that’s only a few centimeters in diameter can thus produce several milliwatts of electricity as long as ambient energy, such as light for solar cells, is available. This energy is collected in a capacitor until enough has been stored to make a radio transmission possible. Because transmitting and receiving requires a lot of electricity — from ten to just under 100 milliwatts — depending on the technology), the network usually remains silent. In other words, the sensors go into stand-by mode — a state in which only a few microwatts of electricity are needed to keep them operating. Every 100 seconds or so, a sensor will “wake up,” take measurements, and communicate with its nearest neighbor. What’s more, they forward information only if this communication reveals something significant — for instance, if all neighboring sensors register higher than normal temperatures.

Such frugal exchanges of data must, however, be reliable. For example, they need to overcome reflections that can block signals or break up data packets. Sollacher is therefore examining sensor nodes with several antennas that can receive radio signals from directions with fewer disruptions. A centrally controlled radio security system that Siemens developed in 2004 for buildings works a little differently. Its network consists of around 15 nodes — for example, smoke detectors — that transmit data from node to node to a base station and then autonomously search for alternative routes if they encounter dead zones.

Siemens also offers processing industries radio-based products, says Kurt Polzer, who is responsible for the development of the wireless field device business at Siemens Industry. “On the one hand,” he says, “switching to radio technology poses risks for an industrial company. For example, if the controller technology fails at a plate glass factory, 1,000 tons of glass can cool down in the melt furnace. But there are also advantages — and you can avoid problems by simply equipping the facility with additional radio sensors. This reduces maintenance costs and increases productivity and production quality.”

Sollacher uses a test network in his lab to study how self-organizing networks function. The network contains up to 80 nodes for measuring temperature, brightness, and humidity. They measure distances to their neighbors in order to pinpoint their own positions, autonomously assign radio channels, and regularly synchronize their internal clocks. They also interpret data — for example, they can determine mean temperatures. They do this by comparing their measurements with those of their neighbors and then using the data to estimate an average value for the overall system. This value even includes measurements made by nodes unknown to them. By exchanging these estimates with their neighbors, they are thus able to quickly calculate the correct mean temperature, which can then be called up from any sensor node.

The network uses a similar process to identify simple patterns — for example, the combination of different measurement values. Sollacher describes the procedure as follows: “Imagine a refrigerated container that is being monitored by temperature, humidity, and door sensors. If the door is closed, the temperature and humidity should be within a certain range. If it’s open, these limits will probably be exceeded.” Each sensor node utilizes predefined ranges, and then uses its actual readings to calculate if all system parameters are within those ranges. Nodes exchange their estimates and the network calculates iteratively whether or not the system as a whole is operating properly. Such functions are crucial for industrial monitoring systems that utilize a large number of sensors.