We elaborate in detail in this section the technical components in hardware cloning, sensor networking, and elevator signal processing. Provided also are our preliminary assessments on the costs of sensor node manufacturing, the success ratio of sensor data delivery, and the accuracy of elevator status inference.
Figure 21. Telos Revision B. Before soldering (1, 2) & after soldering (3, 4)
6.3.1. Cloning Sensor Nodes
Cloning Processing. We choose to clone the ultra low-power sensor node, Telos (See Figure 21). The cloning process involves three steps: (1) PCB manufacturing (2) parts purchasing and (3) part soldering. First of all, the schematic, printed circuit board (PCB) layout, also known as the Gerber file, and bill of materials are open source and available from the TinyOS website. We send the Gerber file to the PCB manufacturer to produce the PCBs from which we receive the printed boards within two weeks.
The most time consuming part is the purchasing step. We need to acquire all the parts we need. Before we know what parts to buy, we spent a significant amount of time studying the parts and the corresponding functionalities in the bill of materials.
The datasheets are carefully examined to make sure that we order the parts with correct footprints. Our attempt purchasing from local suppliers is not successful for the reason that most of the suppliers do not take low quantity orders. We have no choice but to turn to Digikey , a major electronic component distributor that offers a breadth of product lines, provides with online catalogs and accepts low-quantity orders.
After getting all the parts ready, the third step is to solder all the parts on the PCBs.
The components used by Telos are very small and some have special footprints that are almost impossible to solder by hand. We take a stencil and toaster oven approach5.
The idea is to use the stencil to paste the solder paste on the PCB, place the components on the solder paste, and melt the solder paste using a baking oven. A microscope is necessary to check whether the components are well aligned before sending the solderpaste-and-component ready PCB to the baking oven. Temperature control is also important. The components will malfunction if they are overheated for too long.
Figure 22. The illustration of wireless sensor network
Turn-Out Rate. We clone 50 pieces of Telos. After we produce the hardware, install the program, and test, only 5 of them function correctly. Thisnumber is far lower than what we anticipate. Hardware debugging is essentially to identify abnormal voltage level, resistance level and waveform using the electric multi-meter and oscilloscope.
Once the abnormal component is identified, we use the microscope to check the quality of soldering. Two kinds of problems are common: (1) component placement and (2) soldering precision. For example, a 100 ohms resistor is soldered at places for
a 100K ohms resistor. Either too much or too little soldering paste will not be good.
The former creates short circuit over two consecutive pins. In case of the latter, the IC might not be soldered firmly on the PCB. After a round of hardware debugging, we fix most of the malfunctioning Telos and 40 of them function fine at the end.
Problems of the remaining Telos’ are unknown and need substantial rework.
Manufacturing Cost. The manufacturing cost consists of three parts: component, equipment, and labor. The electronic components cost approximately 1600NTD per piece of Telos, and the cost of PCBs is 400NTD per piece. We bought some
equipment for the making and testing. Those include the stencil, microscope, oscilloscope, electric multi-meter, and baking oven. The total cost of these
equipments is approximately 150,000NTD. The final part is the labor cost which is difficult to estimate. We spent time on studying the datasheet, finding supplier, purchasing components, trying out the stencil and toaster oven method, placing components, testing, debugging and fixing. It takes 4 months and 2 graduate-level man power to complete the Telos cloning process. The manufacturing cost is summarized in Table 7 below. The overall cloning cost, 250,000NTD, might not be low. However, going through this cloning process helps us to understand the hardware.
We also gain insights on the manufacturing cost if the sensor nodes will be mass produced later.
Electronic Components 1600/piece NTD Printed Circuit Board 400/piece NTD Material Cost per Piece 2000/piece NTD Total Material Cost 100,000 NTD Total Equipment Cost 150,000 NTD
Table 7. Cost of successfully cloning 40 pieces of Telos
6.3.2. Sensor Network Infrastructure
Elevator. When the elevator arrives at a particular floor, the mote inside the elevator will transmit a message which contains the information of the estimated floor number and moving direction. There are nodes placed on the ceiling nearby the elevator each floor. When the node receives the message from the elevator, it will check the
correctness of the floor number first. If the number is right, the node will transmit the message toward the sink through a wireless sensor network. The node inside the elevator will be the ‘Elevator’ node and the node outside the elevators will be the
‘Source’ node as indicated in Figure 22.
Magnetic Diffusion. The ‘Source’, ‘Relay’, and ‘Sink’ nodes form a wireless sensor network in the building. To route data through the network, we adopt a routing protocol called Magnetic Diffusion. In that, the sink, functioning like the magnet, propagates the magnetic charge to set up the magnetic field. Under the influence of the magnetic field, the sensor data, functioning like the metallic nails, are attracted towards the sink. The magnetic field is established by setting up the proper magnetic charges on the sensor nodes within the range of sink. The strength of the charge is determined by the hop distance to the sink. In Figure 22, the sink node broadcasts interest periodically and then builds a magnetic field upon other source and relay nodes. Once the source node detects elevator door opening and has data (elevator status information) to send to sink, the data will travel from the low to strong charge relay nodes, and finally arrive at the sink which has the strongest charge.
Figure 23. Reachability of individual nodes in wireless sensor network
Data Reachability. For the evaluation, we deployed two sensor nodes with accelerometers in the two elevators and 14 Telos to build the sensor network. The nodes nearby the elevator send the data twice for each elevator arrival for better reachability. The deployment extends for 4 floors, from the fourth to the seventh.
Figure 23 illustrates the placement of the nodes in BL Hall. The duration of
experiment is two hours. During the time, we logged the message received by the sink node.
The reachability of the elevators is shown in Table 8. The reachability of the east elevator is 95%, about 5% higher than the west elevator. This is because the two nodes shown in Figure 23, west 5th floor (W5) and relay center (RC), were out of power during the experiment, and thus cannot transmit messages to the sink. The reachability of the west elevator should be close to that of the east elevator without the failing nodes.
Node ID # Packets # Packets Reachability Sent Received
E. Elevator 60 57 95%
W. Elevator 56 50 89.2857%
Table 8. Reachability of the elevators
6.3.3. Accelerometer Signal Processing
We measure the acceleration of the elevator using the accelerometers to decide the status of the elevator. There are other means to acquire the status of the elevator. For example, one may obtain the elevator status from the elevator itself. Without working with the elevator maintenance, the concern is the safety of the building residents.
Figure 24. Typical accelerometer data pattern of the elevators
Using a camera and broadcasting the video openly on the Web is another option, but there is still privacy concern. The use of accelerometer is less intrusive and we can avoid needing the elevator engineers’ attentions.
Figure 24 shows the typical up-and-down patterns in the acceleration data collected over time. We can identify four phases of operations of the elevator: STILL,
ACCELERATING, MOVING, and BREAKING. A pair of up and down represents
the elevator ACCELERATING and BREAKING. In between such an up and down pair, the elevator is in the MOVING phase. In between the up and down pairs, the elevator is in the STILL phase and it stops and opens its door to let people in and out.
The amount of time the elevator stays in each phase is unique for each floor movement. We collect the acceleration data of the elevator every 0.5s for hours, including periods with heavy and low elevator traffic. There are in total 500 floor movements in 3 hours when collecting the acceleration data. According to the time the elevator spends in two phases, the ACCELERATING and MOVING phases, we can identify well which floor the elevator goes to.
We construct four tables of the maximum values and the minimum values of the accelerating time and the moving time for all-pair movements. When the elevator finishes an operation, the sensor checks these tables to tell which floor the elevator goes to. The system begins on the seventh floor, the initial floor, when the elevator stops and opens the door, the sensor would tell which floor it is and send the status out of the elevator to a sensor network outside the elevator. Our algorithm can identify the floor movement 100%. The four tables also allow us to identify the noises. We create 90 noises by shaking and lightly jumping in the elevator. 90 noises are all detected.