The use of automated vehicle location systems in solid waste collection operations

Transcription

1 The use of automated vehicle location systems in solid waste collection operations CONTACT Bruce G. Wilson, Dept. of Civil Engineering, University of New Brunswick Thuy T.T. Nguyen, Dept. of Civil Engineering, University of New Brunswick Dr. Bruce G. Wilson, P.Eng. Department of Civil Engineering, University of New Brunswick P.O. Box 4400, Fredericton, New Brunswick, Canada, E3B 5A3 Phone: EXECUTIVE SUMMARY In November 2003, the City of Hamilton, Ontario, Canada equipped five solid waste collection trucks with Trimble Telvisant CrossCheck GPRS 1900 Mobile data recorders. Three of the trucks are garbage packers, the fourth truck is a side-loading recycling truck, and the fifth vehicle is a 60/40-split co-collection truck equipped with a cart lifter. These trucks operated in a variety of situations, depending on the day of the week. The Global Positioning System (GPS) units sample the position, speed, and bearing of each vehicle at one-minute intervals. They also record certain events, such as when a vehicle crosses the virtual boundary of a user-defined geofence. This paper summarizes the results of more than three years of experience with this very rich data source. In particular, the paper provides an overview of the use of GPS data in daily collection operations, in routing vehicles on collection routes, in estimating fuel consumption and vehicle emissions, in modeling waste collection operations, and in measuring delays at unloading facilities. Some of the findings from this project challenge years of research into solid waste collection operations. For example, there are literally hundreds of papers on the problem of finding the shortest path through a collection network. However, this research suggests that deviations from the optimal collection route do not have a statistically significant impact on overall collection time for any given route. Conversely, the data clearly show that queuing delays at unloading facilities do add a significant amount of time to the collection day and that these delays may be caused by assigning equal workloads to each collection vehicle. In other cases, the shear volume of data means that statistically valid models of collection operations can now be constructed. INTRODUCTION Historically, municipalities have not collected much data on their solid waste management systems. In the 1980 s, for example, many North American landfill sites did not even have weigh scales to record the weight of material entering the site. Most municipalities do keep records on their collection fleet, but this information is often used only for fleet maintenance and costing purposes. This information is usually collected and stored by operations staff. Solid waste planning staff may not have access to this data or may find that the information that is available is not in a form that is useful to them. The recent development of low cost global positioning systems has tremendous potential to change how municipalities collect data about their waste management systems, the type of data that they will collect, and the uses that they will find for that data. The purpose of this paper is to report on a project in the City of Hamilton, Ontario, Canada to examine potential uses for GPS data from waste collection trucks.

2 In the fall of 2003, the City of Hamilton purchased and installed wireless GPS units on five waste collection vehicles. Two of the trucks are conventional garbage packers operated by city forces. The third truck is a conventional packer operated by a contractor. The fourth truck collects recyclable materials and the fifth vehicle is a 60/40-split co-collection truck equipped with a cart lifter. These trucks operate in a variety of situations, depending on the day of the week. For example, one truck operates in rural areas two days per week, in a suburban setting for two days a week, and in an urban area on the fifth day. All five trucks were equipped with Trimble Telvisant CrossCheck GPRS 1900 Mobile Units. Theses units integrate a GPS receiver with a wireless communications system. The pilot co-collection vehicle was also outfitted with a sensor to indicate when the cart lifter on that truck operates. The data recorder, shown in Figure 1, consists of a small black box mounted under the hood of the vehicle. Every 15 minutes, the GPS data is transmitted over a wireless network to a Web based server. (Source: Figure 1: Trimble Telvisant CrossCheck GPRS 1900 Mobile Unit The GPS units sample the position, speed, and direction of the vehicle at one-minute intervals. They also record certain Events such as when a vehicle s ignition is turned on or off or when a vehicle crosses the boundary of a geofence. Every 15 minutes, the GPS data is transmitted over a wireless network to a Web based server. Each truck was also equipped with a small terminal located on the dash, as shown in Figure 2. Drivers can use this terminal to record a number of pre-programmed messages, which are included in the events database, but few drivers made extensive use of this feature. In addition to the GPS data, the project used weigh scale records for each of the five vehicles and fuel consumption records for the three City owned vehicles. The weigh scale records recorded the weight of material collected by each truck on each load and were cross-referenced to the positional data from the GPS systems using a timestamp. The fuel records are date stamped and include an odometer reading, allowing them to be cross-referenced to the GPS data. The GPS data, the weigh scale records, and the fuel records were combined into a single database to allow for easier manipulation of the data. A final source of data for the project was electronic maps available through the City of Hamilton interactive map website. Among other features, these maps allow the user to measure the length of individual streets and to count the number of households on a particular block face of any street.

3 DATA COLLECTION Figure 2 Message Terminal Mounted on a Vehicle Dashboard Data collection began on Nov. 24, 2003 and continues now. On occasion, small amounts of data (amounting to a few minutes a day) have been lost. These temporary transmission problems appear to occur when the wireless signal is lost because a vehicle is inside a transfer station. Data gaps on the route are rare and even when they occur, they are short and do not pose a problem for data analysis. Accuracy of the positional data is well within the rated accuracy of ±10 metres and can consistently place the vehicles on a specific block face of any street; however the data are not accurate enough to associate vehicle stops with an individual household. The GPS units generate a considerable volume of data. For example, the system recorded 294,214 position records in the first year of operation. The five study vehicles generated an average of about 450 records per truck per day. While all of this data is very useful for this project, the volume of data generated would be difficult to manage should the City decide to equip its entire collection fleet with GPS data recorders. For example, based on the results to date, it is estimated that a fleet of 40 collection vehicles equipped with GPS recorders would generate approximately 4 million data records per year. RESULTS In addition to the obvious uses for the data, such as tracking a vehicle route on any particular day or keeping operational records such as distance travelled and engine time, the GPS data from this project have been used to study a wide range of collection issues. A report on the project listed at least nine potential uses for the data (Wilson 2005) and many of these have been examined in detail. This paper will summarize the work in the following five areas of interest: 1. Measuring Activities and Efficiency on the Collection Route 2. Modelling the Solid Waste Collection Process 3. Estimating On-Route Vehicle Emissions 4. Measuring Delays at Unloading Facilities 5. Measuring Routing Efficiency

4 MEASURING ACTIVITIES AND EFFICIENCY ON THE COLLECTION ROUTE The data can be used to follow the progress of any individual truck on any particular day. For example, Figure 1 shows the route followed by the co-collection truck on a typical Monday as it travels from the garage to a rural collection route, completes the route, and follows a different path back to a transfer station to unload before returning to the garage. Since each position is time stamped, it is possible to calculate the amount of time the vehicle spends on different activities along the route and the distance travelled during each activity. Since the routes are repeated weekly, it is also possible to calculate statistics such as means and variance of time spent on different activities and distance travelled (Agar 2004; Agar et al. 2005; Wilson 2005; Wilson et al. 2007). Collection Route Garage Transfer Station Figure 1. Typical Monday Route Co-Collection Vehicle. An example of the type of analysis possible is presented in Figure 2, which compares the average collection time per household for the different routes serviced by the co-collection vehicle. Collection time per household is largest on the rural Monday route at 46 seconds per household, while collection time per household on the Thursday route is approximately ½ this value, only 28 seconds per household (Wilson 2005). ANOVA analysis showed that collection times on Monday and Tuesday are approximately equal (45 to 46 seconds/household), but there is a statistically significant difference between these routes and the Wednesday and Friday routes (37 seconds/household each). The Thursday route is significantly different from the other routes, requiring only 28 seconds per household. Even this rather crude measure of collection efficiency shows that the time required to collect from a single household in the City of Hamilton can vary significantly from route to route.

5 Collection Time per Household by Route Collection Time (seconds per Household) Monday 2 Tuesday 3 Wednesday Route 4 Thursday 5 Friday Figure 2. Collection Time per Household Classified by Route MODELLING THE SOLID WASTE COLLECTION PROCESS There have been numerous attempts to develop computer models of the collection process over the last forty years (e.g. Quon, Tanaka and Charnes, 1969; Ouano and Frankel, 1976; Everett and Riley 1997; Wilson and Baetz, 2001). The major drawback in all of these studies has been that they were developed based on very limited data collected from time and motion studies or truck logs. As a result, these models have not been very successful in predicting collection times and have not been adopted extensively by waste managers. The quality of the data provided by GPS data collectors provides an opportunity to develop new models of the waste collection process that are based on accurate data and means that statistical techniques can be used to determine the accuracy of the developed models. The data from this project have been used to calibrate a derived probability model (Agar et al. 2005), to develop regression models of the waste collection process (Wilson, 2005), and to develop a new simulation model of the collection process (Nguyen and Wilson, 2007). ESTIMATING ON-ROUTE VEHICLE EMISSIONS The GPS data were combined with the City s records on fuel usage to estimate fuel consumption waste collection vehicles under operating conditions (Agar 2004; Agar et al. 2007). Fuel consumption of these vehicles can be highly variable due to large, but variable, amounts of time spent idling while loading waste into the vehicle. By combining estimates of the amount of time spent idling compared to traveling at normal road speeds, it was possible to estimate fuel consumption and pollutant emissions from the study vehicles. This work found that MSW collection vehicles may be considerably less fuel efficient in the field, and pollutant emissions from those trucks may be higher, than published values for heavy duty diesel trucks might suggest.

6 MEASURING DELAYS AT UNLOADING FACILITIES An obvious use for the GPS data is to identify delays due to specific problems such as traffic, driver breaks, queuing to unload, etc. Delays are indicated by a series of closely spaced positions with time. Although it was originally expected that the data would highlight a number of specific delay areas, some expected delays were not observed, or were observed only rarely. For example, there were few instances of drivers taking extended breaks or rest periods. Many of the drivers worked without breaks or lunch stops. This is likely due to work rules whereby the collection crews are paid on a task and finish basis where they are paid for a minimum time period, even if they finish their route sooner. (Interestingly, many GPS systems are marketed as a method for monitoring drivers for this type of activity.) It was also expected that the data would highlight delays due to traffic bottlenecks. Although some such delays were observed, they were relatively rare and did not tend to repeat week after week. This may be because the drivers are adjusting their routes to avoid delays by taking alternative routes. All of the vehicles are equipped with radios, so drivers caught in traffic would often warn other drivers of potential delays, allowing them to adapt their route. Delays due to poor weather were observed. These delays affected all vehicles and tended to result in a general slowdown in all activities. One area where delays were observed from the data, but not expected, was at transfer stations. Weigh scale records from the City indicated that trucks were weighed in and out of the City s four transfer facilities very quickly and City staff stated that queues at unloading facilities were not a concern to them. However, analyses of the GPS data showed that, not only were there delays at the City s facilities, but also that some of the delays were quite long and that each of the facilities was operating in a different way. This analysis used geofences to divide each transfer station into several distinct activity areas, as shown in Figure 3. Figure 3. Aerial Photo of Transfer Station with Geofences Superimposed

7 At each facility, an area for queuing prior to the weigh scale, an area around the weigh scale, an area after the weigh scale but prior to the tipping floor, and an area around the tipping floor were established. The time spent at each activity was then estimated from the GPS data. The results are presented in Figure 4, which shows the average time spent on each activity at each of the City s four facilities. The results are somewhat surprising. First, they show that the average time spent at one facility, the material recovery facility, was more than 30 minutes per day per truck. Approximately 50% of this time was spent in queue waiting for other trucks to unload. Since this facility is used by approximately 20 vehicles, the queuing delay at this facility alone adds up to about 4 truck hours per day the equivalent of parking an idling vehicle in the yard for half a day. Although delays at other facilities were smaller, they were still significant, amounting to as much as 10 minutes per truck per day. Time (minutes) Mountain Dundas Kenora MRF Facility Wait for Scale On Scale Wait to Dump Dumping Figure 4 Delays at Four City Unloading Facilities Also evident from this analysis was the fact that the delays at different facilities were occurring for different reasons. At one facility, the delay was occurring while trucks were waiting to use the weigh scale. As noted by City staff, this delay was relatively small. At two other facilities, the delays were significant and were occurring after the weigh scale, but before the tipping floor. These delays had not been observed by staff. At the final facility, delays were relatively rare, probably because the facility was under utilized. MEASURING ROUTING EFFICIENCY Throughout this project, several interesting points about individual truck routes were observed. For example, it became clear that some drivers followed the same route from week to week, while others changed the order that they visited different streets from week to week. Secondly, it was observed that even when a vehicle needed to deadhead a street or travel along a street that was not on the collection route, it did not take much time to do so. The question became, did these differences in the route followed lead to measurable differences in the time required to collect waste on the route? Optimal routing of waste collection vehicles has been actively researched for decades, and it is possible to determine the collection route with the shortest length for any neighbourhood. However,

8 until recently it has been impossible to confirm that the driver is actually following the optimal route or to quantify the effect of deviations from the optimal route. The GPS data now allow for the actual route followed by any truck to be compared to the theoretically optimal route. As part of this study, the optimal collection route was determined using TransCAD GIS software and the actual route followed by the truck was tracked using GPS data (Nguyen and Wilson 2007). Then, for each actual collection day, the distance travelled by the truck and the time taken to complete the collection route were determined from the GPS data. For each collection day, the efficiency of collection was defined as follows: ε = Actual distance travelled Ideal distance That is to say that the distance travelled was normalized by dividing by the optimal or minimum route length. Theoretically, an efficiency of less than unity is impossible. In addition, any deviation from an optimal route should result in an efficiency of greater than unity and should require additional time for the vehicle to complete the collection route. Figure 5 presents the results from one collection route. Several interesting observations are evident from this figure. First, it is clear that the truck almost never travelled the same distance from week to week and in some cases the driver drove 50% more distance than the optimal route length. Comparisons of the path followed showed that the driver often changed the route that he followed to complete his route. Even when the driver did follow the same route, there were differences from week to week due to differences in the exact location of turns, etc. The figure also shows that the collection efficiency values are in the range from 1.0 to 1.5 (meaning that on different days the truck travelled anywhere from almost exactly the theoretical minimum distance to 50% more than the theoretical minimum). Despite the variation in travel distances, most of the time, the actual distance travelled was 115% to 125% of the optimal route length. This means that even thought the driver regularly deviated from the optimal route, his total travel distance was close to the optimal value. This suggests that drivers are actually quite adept at finding near-optimal routes through a neighbourhood that they travel regularly. More importantly, Figure 5 shows a least squares regression analysis between collection time and collection efficiency. The results show that there is a very poor correlation (R2 = ) between the time spent on the route and the routing efficiency. The total time required to collect from the route varies between 130 and 230 minutes, a difference of 100 minutes per day. However, very little of this variation (approximately 24 minutes per day) can be explained by the extra distance travelled when deviating from the optimal route. The data suggests that poor routing on this route accounts for an average of 5 to 10 extra minutes on the route per day. It is important to note that the driver was not given an optimal route to follow. He was assigned a collection area and told to collect waste from all households in the area. Thus, by trial and error, the driver has found a route that is only slightly longer than the optimal route and which can be completed in approximately the same amount of time as the optimal route. Theoretically there should be some savings in time and distance if the driver was required to follow the optimal route exactly, however the potential savings do not appear to justify taking control of the route away from a driver that is clearly doing a reasonable job of route finding. In addition, since we do not know the reasons why drivers take different routes on different days, requiring that they strictly follow a prescribed route may not result in any real time or distance savings.

9 Time (min) y = x R 2 = Collection efficiency Figure 5. Relationship between collection time and collection efficiency for the one route. CONCLUSIONS This research has shown that GPS data recorders can provide a tremendous amount of detailed information on the waste collection system that can be used for planning purposes. A number of potential uses have been identified in this report and many other uses are possible. GPS data recorders can provide reliable, accurate information on the position and activities of waste collection vehicles. The data are not accurate enough to associate vehicle stops with an individual household, but they are sufficiently accurate to assign the vehicle to a particular block on a particular street. The GPS data can be used to estimate the time required to perform a number of onroute tasks and to identify delays that occur on the route, a task which until now has been accomplished using time and motion studies. The GPS data recorders generate a considerable volume of data and the volume of data generated is difficult to manage. For example, a fleet of 40 collection vehicles equipped with GPS recorders could generate approximately 4 million data records per year. Fleet managers and solid waste planners need to plan for how they will store and analyze this data. The large volume of data does mean that standard statistical methods can be used to analyze waste collection operations. It is expected that this statistical reliability can be used to improve the modelling of waste collection operations considerably. Some of the findings from this project challenge years of research into solid waste collection operations. For example, there are literally hundreds of papers on the problem of finding the shortest path through a collection network. However, this research suggests that deviations from the optimal collection route do not have a statistically significant impact on overall collection time for any given route. Conversely, the data clearly show that queuing delays at unloading facilities do add a significant amount of time to the collection day. Further experience with, and analysis of, GPS data is expected to provide more insights into waste collection operations.

10 ACKNOWLEDGEMENTS This financial support from the Natural Sciences and Engineering Research Council of Canada, the University of New Brunswick and the City of Hamilton is gratefully acknowledged. REFERENCES Agar, B. (2004). Using GPS to Analyse Municipal Solid Waste Collection. M.Sc.E. Thesis, Department of Civil Engineering, McMaster University, Aug Agar B., Wilson B., Winning A., and Baetz B. (2005). Measuring the performance of solid waste collection vehicles using global positioning systems. 9th Env. Eng. Specialty Conference, Canadian Society for Civil Engineering, Toronto, Canada, June 2-4, CD Proceedings, Paper EV-110. Agar, B., Wilson, B. and Baetz, B.W. (2007). Fuel Consumption, Emissions Estimation, and Emissions Cost Estimates Using Global Positioning Data, J. Air & Waste Management Assn., 57: Everett W. J., Riley P Curbside collection of recyclable material: Simulation of collection activities and estimation of vehicle and labor needs. Journal of the Air & Waste Management Association. Vol47, October Nguyen, T.T. and Wilson, B.G. (2007). Measuring the efficiency of solid waste collection routes, 22 nd Int. Conf. on Solid Waste Tech. & Mangmt, Philadelphia, PA, Mar , 2007, Nguyen, T.T. and Wilson, B.G. (2007). A Simulation Model of Solid Waste Collection Operations Based on GPS Data, 22 nd Int. Conf. on Solid Waste Tech. & Mangmt, Philadelphia, PA, March 18-21, 2007, Ouano, E.A.R. and Frankel, R.J Decision criteria in solid waste collection. J. Env. Engrg. Div.. Vol 102(EE3) Quon, J.E., Tanaka, M., and Charnes, A Simulation model of refuse collection policies. J. Sanit. Engrg. Div., ASCE, vol 95 (3): Wilson B.G., Baetz B a. Modeling municipal solid waste collection systems using derived probability distributions. I. Model development. Journal of environmental engineering, pp Wilson, B.G., Agar, B. and Baetz, B.W. (2004). Calibrating a Solid Waste Collection Model Using Electronically Collected Data, 19 th Int. Conf. on Solid Waste Tech. & Mangmt, Philadelphia, PA, March 21-24, Wilson B.G. (2005). Collection and Analysis of Detailed Solid Waste Collection Data Using Onboard Global Positioning Systems. Final report submitted to the City of Hamilton. Department of Civil Engineering, University of New Brunswick, Fredericton, New Brunswick, Canada. Wilson, B.G. and Vincent, J. (2006). Estimating Waste Transfer Station Delays Using GPS, Article submitted to Waste Management Journal. Wilson, B.G., Agar, B.J., Baetz, B.W., and Winning, A. (2007) Practical applications for global positioning system data from solid waste collection vehicles, Can. J. Civil Eng., 34(5):