Issue 2/2008
03/02/08
Performance and Comfort for Elevators
Beat De Coi, Daniel Lippuner and Peter Nebiker
Nowadays emerging camera technologies facilitate different new applications in elevator technology, which enhance the elevator performance, security, and comfort. For the elevator’s performance optimization the level of occupancy in the car is a key figure. Existing overload sensing devices measure the weight of the elevator carload for safety reason but cannot detect the occupancy level of the car. For security applications a presence detection of persons and objects in the car is needed. Standard intrusion sensors are not applicable since they don’t detect non-moving objects. This paper describes new low cost vision systems based on standard CMOS camera technology, which provide solutions to elevator’s performance and security applications. A possible application to enhance to passenger’s comfort is shown. The new camera based sensor technology is explained and the different applications are discussed.
Category: Issue 2/2008
Posted by: Editor
Since the elevator was invented more than hundred years ago, sensor technology was used mainly to control the elevator and elevator doors. Some examples for sensors are hoistway information systems by using mechanical, magnetic, or optical switches, rotary encoders to read the angle position of the motor shaft or, for passenger safety reasons, mechanical, capacitive, or optical doors safety switches. Also touch sensitive sensors have been introduced to be used in fixtures. However, the use of sensors for security reasons or to reduce waiting time was very limited in the past. One of the reasons that such applications could not become a commodity status until today is the complexity of the application. Conventional sensor technology fails, e.g. to determine the number of persons in an elevator cab. The key to penetrate into new areas is the usage of information technology concepts. Information technology deals – like the word says – with information. However, conventional sensor technology produces very limited information. E.g. a light barrier can only provide a digital signal, namely interrupted or not interrupted. Even if we study the information potential of a light curtain, we end up maybe with 50 individual digital signals. This is way too less in order to solve problems like ‘is the car empty?’ (no person in the car, no letter bomb inside etc.) or ‘how many persons are entering a car?’
The use of digital semiconductor cameras open up a new dimension when they are used as sensors. If we are using a very simple color VGA Webcam chip with a resolution of 640 by 480 pixels, a total of 307,200 data points can be used to evaluate the image. If we take into consideration that each pixel does not only contain the information ‘yes’ or ‘no’ but 3 colors with a grey scale each of 10 bits (1024 different grey levels per color), the total amount of data is 943,718,400 bits. Or, in other words, 944 Mega bits. Because most of the desired applications need real time capability, at least 10 to 20 images per second have to be processed. It speaks for itself, that only pure application of information technology helps to analyze the images in order to provide information like how many persons enter an elevator.
This paper describes the function principle of camera-based sensors and provides an efficiency analysis for a sensor that measures the occupied floor space in an elevator car.
Functional Principle of a camera-based sensor
A camera-based sensor consists typically of five building blocks:
Camera Chip
This part captures the image that is depicted by the camera lens. It contains photosensitive picture elements (pixels) that transform the received photons into electrons and store them (the charge of the electron) in a capacitor. Once an image is captured, which means that all pixels in the pixel array are illuminated simultaneously, the stored charge of every pixel is converted to a digital word and is read out pixel by pixel by the CPU. Thus, the illumination process is parallel for all pixels but the read out procedure is serial pixel by pixel. Hence, the readout process is rather time consuming because of thousands of pixels have to be addressed and read out individually. Without any sophisticated information technology concepts, this process would last for seconds or minutes, which would not be acceptable. The technologies to do so are available due to the advanced developments in the consumer camera business.
Central Processing Unit CPU
This is the main block that does all the control of the camera as well as the evaluation of the image data, which is described in the Program Memory (Software) section. The CPU controls the illumination time of the camera chip in order to get pictures that are not under- and also not over-exposed. The CPU also controls the data flow between the camera chip, the program memory and the image memory and it provides the output signal via the interface to the elevator or door control, or the user if desired. Because the evaluation of images requires extremely high computing power, the CPUs usually are not simple microprocessors but digital signal processors (DSP). These processors ore optimized to process hundreds of millions of mathematical operations in a second.
Image Memory
This part contains the last captured image as well as one or more reference im ages. It is a random access read and write memory, also called RAM. The process of image evaluation is generally speaking a mathematical transformation of the actual image and a comparison with a previously taken or a reference image. Let us consider the following process: First, a reference image is taken and stored into the reference image memory. Afterwards, all images taken are compared with the stored image pixel by pixel. According to the example shown in Figure 2, all pixels that deviate by a value of more than 4 are set to true. The result in the example shown is a 5-pixel deviation. A simple pixel counter sensor triggers its output for example if there are more than a given number of pixels with the status ‘true’.
Program Memory (Software)
This part contains the software that runs the CPU in order to control the whole sensor as well as the algorithms that evaluate the images. This part is usually stored into a non-volatile flash memory.
Interface
The interface is the part that submits the results evaluated by the CPU to the elevator or door control device. It can be a simple relay that opens or closes a contact if a specific image condition is detected. It can also be much more sophisticated, e.g. a CAN-Bus that submits how many passengers are in an elevator car. The type of interface is dependent on the application as well as on the requirements of the elevator or door control unit.
Applications
Sensors, based on the above principle can be used for various applications. One of them is the determination, how much floor space of an elevator cab is occupied. If this measurement is above a given threshold, an output relay is triggered that instructs the elevator control not to collect subsequent hall calls due to limited space in the car. This application is discussed in detail in the section 4, efficiency analysis.
Another application is to determine whether an elevator car is empty or not. Such a sensor can be used for security reasons. Assume a penthouse that is served by an elevator, e.g. from a parking garage. To access the penthouse by the elevator, a room key is necessary. However, a burglar could wait in the elevator as long as somebody in the penthouse makes an elevator call. Certainly, the inhabitants would be sure that nobody is in the elevator when they call the elevator, which can be done with a camera-based sensor. Such a sensor has also the capability to detect a bomb in the elevator car. The following pictures illustrate the function of this type of sensor:
Once a camera-based sensor is installed in the car it is also possible – with the appropriate image-processing algorithm – to count the number of people in the elevator. This is especially helpful when a destination dispatcher manages the elevators. Very often, only one person enters the destination floor number and then a group of others follow this person without entering their floor number, because they want to go to the same floor. The biggest advantage of the destination control dispatcher is eliminated because no accurate information about the number of persons traveling together is available. Hence, a sensor that counts the number of passengers entering or leaving the cab is essential for this kind of elevator dispatchers.
Another application is to count the people standing in the lobby waiting for an elevator. A crowded lobby can be used as an indication that an elevator will be called soon. These sensors can also be used in VIP floors in order to increase the service performance for that floor.
Since camera based sensors usually work with the environmental light, i.e. the sunlight, the car lighting or the lobby lighting, there are certain restrictions in the applications of the vision sensors. It is essential to have an appropriate lighting for the proper function of the sensors. Fast changing illumination of the scenes, e.g. in panoramic elevators, may prevent the sensor from a reliable function, too. Indeed, this problem may be overcome in the future with more sophisticated image processing algorithms, which in addition to the changes in the gray levels also take into account the edges in the scenery. However, the only way to eliminate this limitation is to use real 3D-cameras that do not only provide grey-scale orcolor information, but also distance information per pixel. If such devices are based on the Time-of-Flight principle, they work completely independent of the naturally given light by the environment.
Efficiency Analysis
Let us present an estimate of how the application of an ESPROS/VOL sensor may considerably increase the efficiency of an elevator system. We will do this by deriving an average upper bound for the time gain per transported person as a function of the number of floors and a few elevator-specific parameters. After the description of the investigated scenario we make a couple of (reasonable) numerical assumptions. Based on the assumed parameter values we derive a few intermediate results and present simulations on the improved elevator capacity.
Considered Scenario
A typical situation, where the elevator efficiency could benefit from an intelligent volume sensor, can be found, e.g., in a typical business hotel around the morning checkout time. The hotel guests have appointments at a given time so they have to be somewhere on time. The time window of this kind of ‘rush-hour’ is rather small. As a consequence, the elevator has to cope with a person flow on each floor intending to reach the hotel lobby as soon as possible. The task of the volume sensor is to prevent needless elevator stops in case of a full elevator car. It is to mention that, in the situation of this morning down peak, the car is fully occupied before the maximum number of passengers is reached because the passengers carry their luggage that needs floor space too.
Assumptions
For simplicity we assume a uniformly distributed person flow P . [Persons/second] on each floor. Furthermore an up-stream of persons (from the lobby back to the hotel rooms) is neglected in this example. After the cabin is emptied in the lobby the elevator control chooses the floor with the largest (and probably also the most impatient) crowd in front of the landing door. In the following table the parameters are listed.
The above-described system was implemented as a state machine in a Matlab environment. The simulated elevator position and the cabin-full indicator signal (as it would be determined by the ESPROS /VOL sensor) is visualized in Figure 4 as a function of time.
We may point out the following aspects:
- During the start-up phase of the simulation the elevator capacity is not fully exploited, hence the performance improvement of an installed ESPROS/VOL sensor is only marginal.
- The constant person flow on each floor will lead to a permanently (over-) loaded elevator cabin, such that the ESPROS/ VOL sensor can remarkably reduce the average transport time per person, i.e. increase the efficiency of the elevator system.
- After the start-up phase one can recognize a periodic pattern in both simulations (with and without ESPROS/VOL sensor). Roughly speaking the elevator periodically visits each floor in the fairest possible manner.
To derive an easy-to-handle formula for the average upper time gain per transported person we simplify the periodic pattern according to the approximate profiles shown in the figure below. We call it an “upper” time gain, since it considers the “worst case”, where the cabin capacity is completely filled out by persons of one floor only. This worst case may also be considered as a part of a larger elevator- position profile, where, e.g. only the lowest N floors of a very high building encounter this situation of completely filling out the remaining cabin capacity left by the floors higher than N. The time gain formula will be derived in the following subsections.
It is clearly visible that the average waiting time of the passengers is reduced by the time to stop the car unnecessarily (decelerate - open the door - close the door - accelerate). The following analysis provides a mathematical solution in order to determine the waiting time reduction.
Elapsed Time Between Two Stops
Given the maximum speed and acceleration, the elapsed time between two elevator stops, which are H [meters] away from each other, shall be derived. The velocity of the acceleration phase, given by ν(t) = ν• • t is being increased up to the maximum speed ν. Hence, the speed-up time lasts t1 = ν/ν• [seconds] over a distance of x1 = ν• • t1²/2 = 0.5• ν²/ ν• [meters]. If we assume that the deceleration phase is the time-reverse analogon of the acceleration phase, the constant-speed distance, described by x2 = H – 2x1 [meters], is passed within t2 = x2 /ν [seconds]. Note that if x2 is negative, the distance H is too short for the elevator to accelerate to the maximum speed. Collecting the above intermediate results the overall elapsed time between two elevator stops yields
For clarity we restrict ourselves to the case tH = H / v + v / ν• , where the elevator manages to accelerate to the maximum speed.
Average Time per Transported Person (Case Without ESPROS/VOL)
We next calculate the totally elapsed time of one pattern period. It consists of the accumulated time spans of the upward and downward motions, tUP and tDOWN, respectively, and the intermediate serviceable and redundant time spans, tIO and tRED, respectively. The former two can be obtained according to
and
Since the number of serviceable stops is 2N, the corresponding time span yields
During the downward motions, the elevator redundantly stops 1 + 2 + … + N-1 times, hence
The total number of transported persons during one pattern period amounts to – independently of the existence of an ESPROS /VOL sensor – N times PE. So the total average time per transported person in the case of a missing volume sensor is given by
Average Time per Transported Person (Case With ESPROS/VOL)
The accumulated time spans of the upward motions tUP are identical to the case without volume sensor. This holds also for the sum of the serviceable time spans tIO. The ESPROS/ VOL sensor, however, completely prevents redundant elevator stops, such that tRED.=0. Moreover, it reduces the sum of the downward-motion times, which is identical to the upward- motion time, i.e. tUP = tDOWN. Hence, the total average time per transported person in the case of an installed volume sensor results in
Average Time Gain per Transported Person using ESPROS/VOL
By taking the difference of eqns. (6) and (7), we obtain an upper limit of the average time gain per transported person achieved by the ESPROS/VOL sensor, i.e.
It makes sense, that the time gain is remarkably high for buildings with many floors (N), for small elevator capacities (PE), for small acceleration values (ν•), and for large opening/ closing times of the cabin doors (tD). A high elevator speed (ν) increases the time gain as well, because it directly improves the efficiency of the direct downward motions.
The simulated average time spans (per PE persons, for better visibility) for different person flows are depicted in Figure 6. The vertical axis shows the time needed to bring a fully occupied car down to the lobby level (Minutes per elevator load). The dotted curves are simply obtained by subtracting the theoretical time gain (8) from the simulated time spans without volume sensor (thin solid lines). As a consequence, we expect the simulated time spans with volume sensor (bold solid lines) to be situated somewhere in-between. This is indeed the case for all subplots in Figure 6.
It is not surprising that – for low personflow rates – no (or only marginal) improvements can be achieved by means of a volume sensor (e.g., for ν• = 1/60 [persons per second per floor]). However, we also note that the theoretically achievable time gain can really be reached exactly for high enough person-flow rates (e.g., for ν• = 1/10 [persons per second per floor]).
Waiting Time Reduction
In the discussion above we have derived a formula for the time gain from the viewpoint of the elevator system rather than from the passenger’s viewpoint. In this paper we restrict ourselves to numerically evaluate the waiting time reduction on the basis of the following realistic example. We assume to be in a typical business hotel consisting of 2 elevators with aspeed of 2.5 m/s and a capacity of 8 passengers each, one lobby level, 10 room floors, 50 rooms per floor (90 % of them are occupied), people carrying luggage that uses the same floor space as the person itself, a down peak at 8:00 am with a Gaussian distribution and a standard deviation of 30 minutes.
The above example was also implemented as a state machine in Matlab. Due to the two elevators and the timevarying stochastic person flows, its implementation is more complex than the first example. The simulation results are plotted in Figure 7. The ESPROS/ VOL sensor can reduce the final average waiting time by 54 % (i.e. from 101 seconds without to 46 seconds with volume sensor).
Conclusion
The morning down peak in a typical business hotel is a good example to show the remarkable increase of elevator performance by using sensors that measure the occupied car floor space. The same level of performance improvement cannot be achieved by load weighting or people counting devices since they do not take the carry-on luggage of the passengers into consideration. The use of camera-based sensors allows deriving new control signals in order to reduce passenger-waiting time and to make elevators safer. The products ESPROS/ VOL, ESPROS/LOB, and ESPROS/SEC are examples of such sensors that are easy to use and simple to integrate also in already existing installations in order to improve elevator efficiency.
The study presented in this paper provides information to estimate the passenger time reduction by using sensors, which determine the occupied car floor space. A highest gain can be achieved in high buildings (with ten or more floors), with a high car speed (1.6 m/s and higher), slow doors with long open hold time, and small cabs.
References
Edward A. Donoghue: A17.1 Handbook Safety Code for Elevators and Escalators, 2000 Edition. The American Society of Mechanical Engineers. Matlab, high-level language for all kinds of technical computations: www.mathworks.com
Authors Information
Beat De Coi, born 1957 in Switzerland completed an apprenticeship as a mechanical designer in Switzerland. He developed mechanical parts for military radio systems where he became familiar with the design of products for very harsh environments. After this first industry experience he continued his studies and he received a BSc. In Electronics (Dipl.-Ing. FH) in 1984 from Juventus Schools of Zurich. In 1986 he founded CEDES AG to develop and manufacture optoelectronic sensors. In 1992, he made an Executive MSc. in Operations Management and Logistics at the Graduate School of Business Administration of Zurich. In 1998, De Coi was nominated as “Entrepreneur of the Year” in Switzerland and in 1999 he received the KMU-Oscar; an award for the most innovative entrepreneur. He is President and CEO of the CEDES Group.
Daniel Lippuner, born 1970 in Switzerland, studied electrical engineering in the Swiss Federal Institute of Technology (ETH) in Zurich (1996). There he also finished his Ph.D. in the fi eld of adaptive digital signal processing and Kalman filtering (2001). He then worked for three years as an engineer in a technology-consulting group at Siemens Switzerland Ltd. At the beginning of March 2005, he started his new work as a team manager (for optomechanics) at CEDES AG (business unit elevators).
Peter Nebiker, born 1967 in Switzerland, graduated in Physics from the Swiss Federal Institute of Technology (ETH) in Zurich. He was working for several years in the private research institute “Paul Scherrer Institute” in Villigen, Switzerland and received his Ph.D. degree in semiconductor physics in 1992. From 1992 until 2004 he was working with Siemens Building Technologies Ltd., Zurich, in different positions as product line manager for gas warning systems and head of R&D for fire detection. In 2004 he has joined the CEDES group where he is responsible for the business unit elevators.
2/2008
