MPC
Research Projects (2009-10)
Identifying Number
MPC-323
Project Title
Risk-based Advisory Prevention System for Commercial Trucks under Hazardous Conditions
University
Colorado State UniversityProject Investigator
Suren Chen
E-mail: suren.chen@colostate.edu
Description of Project Abstract
In the United States as well as other developed countries, road accidents are causing more injuries and casualties than any other natural or man-made hazard. Truck drivers merit special attention not only because of their numbers (approximately 2.8 million in the U.S.), but also because they face huge safety risk from traffic accidents- 7 times more likely to die and 2.5 times more likely to suffer an injury than was the average worker (NIOSH 2007). Large commercial trucks are especially vulnerable to single-vehicle accidents (e.g. rollover, sideslip) under hazardous driving environments on rural highways. The hazardous driving environments include inclement weather (e.g. strong crosswind gusts, snow, rain, or ice) and/or complicated terrain (e.g. steep slopes or sharp curves) (USDOT 2005). Each year in the United States, adverse weather alone is associated with more than 1.5 million vehicular crashes, which result in 800,000 injuries and 7,000 fatalities (The National Academies 2006). The traffic safety problem under hazardous driving environment is especially critical for large commercial trucks due to the fact that these vehicles often have to be operated under adverse environments in a pretty high driving speed.
Across the whole United States and other areas in the world, hazardous winds, together with other adverse weather and adverse topographic conditions, have been blamed for many single-vehicle crashes every year, especially those involving high-sided trucks (Summerfield and Kosior 2001; Saiidi and Maragalas 1995; Edwards 1996; Baker 1991; Young and Liesman 2007a, b; Chen and Cai 2004; Chen et al. 2008a). Due to the unique site-specific coupling nature between vehicle, infrastructure and environment which varies with time and from one location to another, these challenges can not be well answered with existing prevention technologies. Therefore, there is a critical need to develop a realistic study on risk-based prevention strategy, which can be applied to different vehicles, different drivers, at different locations and under different hazardous environments based on vehicle-specific realizations of safety risk.
Project Objectives
The objective in this proposal is to provide customized risk-based prevention information for commercial trucks under site-specific hazardous environments. A critical need for such a study is evidenced by significantly higher safety threats faced by large-truck drivers under adverse driving environments than normal conditions, the lack of a reliable and adaptive advisory prevention model, and continuing threats faced by thousands of truck drivers everyday.
Project Approaches/Methods
The proposed study will develop a prevention system which can provide customized advisory driving speeds and decisions for each specific type of truck and under each site-specific hazardous condition. In order to do that, historical accident data for two representative states in past ten years will be used for both analysis and validation purposes. Neural network technology will also be adopted in order to provide accurate and prompt advisory information. The detailed research plan is introduced as follows:
Task 1. Historical accident data analysis for critical variables
Highway Safety Information System (HSIS) (http://www.hsisinfo.org/) is a database sponsored by Federal Highway Administration (FHWA) and has detailed traffic crash data from nine states across the United States. The 10-year detailed crash data on major interstate highways, US highways and state highways in both Utah and Illinois has already been acquired by the PI for the proposed study. Utah and Illinois, as a mountainous state and a northern state respectively, are ideal locations for the proposed study due to their typical adverse driving environments, such as strong winds, ice and snow-covered road surface and complicated terrain, etc. Besides, these two states are those with the most comprehensive accident and injury data among the total nine participating states in the HSIS database. All the datasets are in SAS format and Table 1 gives the primary variables to be analyzed in this task, which are introduced as following three categories in detail:
The crash data will be analyzed using cross-tabulations and modeling such as binary logit and ordered probit models to correlate the environmental variables (as shown in Table 1) and injury data to find critical variables from the perspective of accident and injury severity. Firstly, the most vulnerable truck types will be identified based on the vehicle type definitions and the cumulative percentage involved in the single-vehicle crashes. Secondly, the correlation analysis of these variables to the "contributing factors", "driver injury severity" and "type of collision" will be conducted. As a result, those factors from the "environmental variables" and "vehicle/crash variables" in Table 1 which are critical to injuries from single-vehicle crashes will be identified. The 10-year datasets of those critical factors will be further analyzed to get probabilistic distributions of each factor. These results will be used to consider uncertainties in the reliability-based traffic crash risk prediction model in Task 2.
Task 2 Reliability-based Crash and Injury Risk Analysis
It is well known that the uncertainties of the highway infrastructure and vehicle properties, such as road surface friction coefficients, geometric parameters and mechanical system properties of the vehicles may fluctuate in the vicinity of the nominal values as random variables. These critical factors and their statistical distributions obtained from Task 1 will be brought into the existing "deterministic vehicle single-vehicle crash dynamic simulation model" developed by the PI to consider uncertainties. Reliability theory (Ang and Tang 1975), which has been used widely to consider uncertainties and risks in many fields, will be adopted to derive the risk level of crash occurrence for each crash type. With the reliability-based model, the crash risk (i.e. probability of occurrence) of each crash type (e.g. rollover on- (or off-) road and sideslip on- (or off-) road) will be identified, which finally establishes the relationship between "site-specific environment" and "crash risk/type". The simulation model will be calibrated with actual crash cases from the remaining 25% of actual crash data. As a result, crash risk of one particular truck under particular driving conditions can be rationally predicted.
Task 3. Advisory Safety Prevention Based on Desired Reliability Index
As discussed earlier, appropriate advisory driving speed and decision of traffic closure is very critical on reducing sing-vehicle crashes and associated injury on rural highways. Therefore, the prevention task will aim at solving an inverse problem of the risk prediction model established in Task 2: try to find the appropriate advisory driving speed/decision which can achieve the desired target safety reliability level. The desired target safety reliability level is set by the users, such as policy makers, trucking company, agency officials or individual drivers. Extensive simulations of risk prediction of a comprehensive coverage of possible combinations of vehicle, driving and environmental conditions will firstly be conducted based on the calibrated risk-based analytical model in Task 2.
With the simulation results, a typical back propagation artificial neural network (ANN) with one or two hidden layers will be trained for the prevention purpose (Mehrota et al. 1996). As shown in Fig. 1, the input layer primarily includes driving environmental information, such as weather (e.g. wind speed, ice, snow and road roughness) and topology information (e.g. horizontal curvature, camber angle and slope angle), vehicle information (e.g. type, weight, load percentage) and the desired safety reliability risk index. The output layer will be driving information, such as advisory driving speeds and/or decision of closing traffic. The training data of the ANN will be primarily from the simulation results from the reliability-based risk prediction model in Task 2 and the 75% historical data used in Task1. The ANN model will be validated with new sets of simulation results from the model in Task 2 and also the remaining 25% historical data. Once the acceptable accuracy is achieved, the training of ANN is successful and the ANN model will become ready to be used for prevention prediction.
MPC Critical Issues Addressed by the Research
- High-risk rural roads
- Rural transportation operations
- Effective Safety Management
- Human Factors
- Heavy vehicles and commercial trucks.
Contributions/Potentials Applications of Research
Advisory information such as advisory driving speeds, usually decided by the transportation agencies, is typically uniform for all the vehicles of the same type even under essentially different driving conditions. Under different hazardous driving conditions, there will be different optimal advisory information corresponding to each risk level. The proposed methodological framework of the vehicle-specific mitigation provides customized advisory prevention information to effectively reduce the risk to the user-defined target level. The customized advisory information also lays critical background for the "performance-based" or "consequence-based" traffic safety design and prevention for highways. By providing the vehicle-specific advisory mitigation information, a consistent level of the target safety reliability index among different trucks, at different locations or under different driving environments, can be achieved. Because the proposed model is the critical basis of many other prevention studies under adverse environments, opportunities for long-term research, such as reducing non-recurrent congestion, ITS and emergency management are created for future explorations. Based on this study, a more comprehensive study will be actively pursued to seek federal funding on rural highway safety.
Technology Transfer Activities
Once the research is finished, the results can easily be transferred to software which can be used by many stakeholders. Through communicating with some stakeholders, the PI believes the prospective of technology transfer of this study is very promising.
Time Duration
July 1, 2009 through June 30, 2010
Total Project Cost
$68,710.00
MPC Funds Requested
$31,224.00
TRB Keywords
Commercial truck; accident; hazardous environment; prevention
References
(not shown because of space limits)
