an unknown reason, e。g。 a hydraulic hose rupture that caused a dozer to burn as the fluid was sprayed on the hot engine, (2) unsafe acts and human related events, e。g。 a worker’s mistake in providing nitrogen instead of oxygen asphyxiated his coworker, which are mostly addressed by the safety training provisions, and (3) com- municational and managerial errors。 The remaining data could be either those contributing factors that linked the accident to design with regard to the model of Behm (2005) or environmental condi- tions that have made the accident more likely or more serious。 Reese and Eidson (2006) believe that failure to consider the sur- rounding factors will let the root causes remain uncovered。

The synopsis of an accident analysis can better demonstrate the collection and refinement procedure。 In March 2004, a foreman drove a 16 year old framing helper home, whom had fallen 10 feet hitting his head on concrete, and thought aspirin would suffice to alleviate the boy’s severe headache。 Five hours later, observing signs of a traumatic brain injury, the boy’s father telephoned the foreman to drive him to a hospital, where he lost his life one hour later。 Among the main causes of the incident, this methodology fil- ters out the violation of child labor laws, language difficulties because the victim is Hispanic, and the foreman’s failure to sum- mon timely medical services, and in return, focuses on the unpro- tected scaffold which links the accident to design, as well as the height of the platform and the hard pavement which made the consequences more severe。

2。2。Data categorization

In the next step, the refined data should be used to distill and express the risk drivers based on the criteria defined in the theory section。 A most important criterion is to select those drivers that are measurable by available measures of interest, that is, in this research, Building Information Models and model checkers。 Build- ing Information Models normally contain architectural, structural, mechanical, electrical and piping features of a building, as well as some environmental information such as positional  changes  of the Sun and terrain, and in more advanced versions, they may con- tain project schedules, material takeoffs and machinery。 Depend- ing on the Level of Design (LOD), these features will come with additional attached information that makes further analyses possi- ble。 Risk drivers found in the accident investigations are a number of rules to be checked, from a simple test to confirm the presence of an object to complicated numerical computations。 Solihin and Eastman (2015) distinguish four classes of rules that are currently being discussed in the realm of automatic model checking, based on the complexity and the data they need to be processed。 Fig。 1 demonstrates how this research exploits Solihin and Eastman’s classification to select the drivers that potentially can be detected automatically at the design  phase。

In as much as BIM technologies will not stop evolving, this clas- sification helps to dispel the worry that not all the  measures needed to detect these drivers are currently ready to use。 Each class of rules requires more complex processes than their prede- cessors。 As shown in Fig。 1, if a rule applies to a set of components and data available in a building model, it will be categorized into one of the four classes, otherwise, the next driver rule will be examined。 The second decision block  of  Fig。  1  checks  whether the rule requires any calculations to direct it to the next decision block。 Anything that asks for an explicit attribute or existence with no further calculation will remain and will be placed into class 1。 Solihin and Eastman mention that simple calculations are not yet embedded in the models, but they can be either asked from the user or done by an elementary programming engine。 This is the subject of the third decision block。 If answered no, these simple calculations will form class 2。 Otherwise, it means that there are some  questions  with  such  extended  or  advanced    computations