The design of a human reliability assessment method for structural engineering

In the recent past a number of buildings collapsed in the Netherlands under apparent normal circumstances. The causes of these failures are predominantly human error within the design or construction of the building. Examples of this are the collapse of five balconies of an apartment building in Maastricht in 2003, and the partial collapse of a roof structure under construction of a football stadium in Enschede in 2012. Based on these developments it is of importance to investigate the current building practice concerning the occurrence of human error. The objective of this research is to investigate the effect of human error within the design process on the reliability of building structures. Based on this, the following research question is defined: What are the consequences of human error within the structural design process on the structural reliability of a typical building structure? The research question is answered by proposing a Human Reliability Assessment method and subsequently analyse the effect of selected human actions within the structural design process. This method is envisioned as a monitoring method for use within engineering/construction organizations. The research consists of two consecutive parts. Firstly a literature study is performed to examine the current knowledge concerning human error in structural engineering. Secondly, based on the literature findings, a model for Human Reliability Assessment in structural engineering processes is proposed. This model is subsequently used to investigate the effect of human error within a specified structural design process. literature study The literature study focusses on four aspects: the occurrence of structural failure, the basic aspects of human error, the basics of Human Reliability Assessments and probabilistic quantification methods. Concerning the occurrence of structural failure, it can be concluded that the majority of the failures are caused by human error (Fruhwald, Serrano, Toratti, Emilsson & Thelandersson, 2007). In most researches a value of eighty to ninety percent is mentioned (Ellingwood, 1987; Stewart, 1993; Vrouwenvelder, 2011). Based on the researches of Fruhwald et al. (2007), Boot (2010) and ABC-meldpunt (2011) it can be concluded that the occurrence or errors are of the same order of magnitude for design and construction, with slightly higher frequencies for the design phase. An important aspect of failure is that in general multiple causes can be identified (CUR, 2010), and that taking away one of these causes usually mitigates the undesired situation. A useful model to represent error causation is the “Swiss cheese“ model (Reason, 2000; Reason, Carthey & de Leval, iii 2001). The model exists of several defensive layers between an hazard and an undesired situation. In an ideal world these layers would be intact. However in the real world holes are occurring, making an undesired situation possible. Another relevant aspect of failure is the cognitive level on which an error is made. A subdivision of this is given by Reason (1990): a skillbased level, rule-based level and knowledge-based level. This subdivision is roughly based on the complexity of the task at hand and the level of attention. One method to investigate human error within design is by means of a Human Reliability Assessment (HRA). These techniques mostly contain three basic techniques (Kirwan, 1994): identify which errors can occur, deciding how likely the errors are to occur and reducing this error likelihood. Most of the HRA techniques are aimed towards subdividing a process in a task sequence, and subsequently analyse these task sequences on human error. An example is the ‘Cognitive Reliability and Error Analysis Method‘ (CREAM), which is used within the main research. The last aspect discussed in the literature study is the use of probability analysis techniques for quantifying human error probabilities. A frequently used technique is reliability analysis methods which focus on relative effect of failures on the global reliability index of the structure. Another technique is scenario analysis, in which scenarios for errors are investigated to quantify relative consequences associated with these errors. A useful computation method for these kinds of analysis is Monte Carlo analysis, which uses repeated random sampling to calculate results for the analysis. main research In order to investigate the effect of human error in design tasks, a HRA method for specific use within engineering tasks is proposed. A simplified flow chart of this methodology is presented in figure 1. The model encompasses basically four elements: A qualitative analysis, a human error quantification stage, a design simulation stage and a probabilistic analysis. Qualitative Analysis Human error quantification Design simulation Probabilistic analysis Kwalitatieve analyse Menselijke fout kwantificatie Ontwerp simulatie Probabilistische analyse Identify considered process Select scenarios to be analysed Identify context Identify design steps Design steps overview Figure 1: Basic steps within the HRA model The first step in the HRA model is to define the process of interest and its boundaries (qualitative analysis). Furthermore, a selection of the most error prone processes within the overall process is required in order to focus the HRA efforts. The selected process is a structural design process of a beam element within a common office building. The office building is envisioned as a framework of concrete beams and columns supporting a slab floor. The overall stability is arranged by means of a concrete core. Within the analysis two beam types are considered: a statical determined beam element iv and a statical undetermined beam element. Furthermore two scenarios for specific analysis are selected: the level of professional knowledge and the level of design control. The second step within the HRA method is to quantify the probability of failure within an individual design task. This probability of failure is represented by a probability distribution function expressed by two parameters: a Human Error Probability (HEP) and an Error Magnitude (EM). The EM is a parameter which describes the severity of an error. The procedure for determining HEPs consists of two methods: a basic HEP method and an extended HEP method. The extended method is labour intensive and requires quite some knowledge concerning human factors. The simplified method requires considerate less efforts and knowledge, however this method is only applicable for standard design tasks. The simplified method distinct seven basic design tasks, each subdivided in three cognitive levels: a rule-, a skill- and a knowledge based task level. The third step is to combine the task probability distributions to obtain an overall probability distribution of the element strength due to errors in the process. For this, a Monte Carlo simulation procedure is proposed. Within this simulation process, each design task is modelled with an algorithm which models the design task at hand and the occurrence of failure. Furthermore design control is modelled as well in order to investigate the proposed scenarios. For this a subdivision is made between self-checking (by the designer) and normal supervision. Based on the analysis performed in the case study it can be concluded that the proposed simulation method is useful for combining task probability distributions into an overall probability distribution. However improvements are required for practical use of the model. The last step in the model is to determine the probability of failure of the engineered structure. For this a probabilistic analysis method based on plastic limit state analysis is proposed. The overall probability distributions found in step three combined with probabilistic loading conditions are used to determine the structural failure probability. Based on the analysis is can be concluded that the structural failure probability can increase considerable. Finally it can be concluded that the proposed HRA model has the potential to quantify the effect of human error within carefully defined boundary conditions. However further research is required to increase the accuracy of the model and its practical use. From the case study it can be concluded that the statical determined beam element is slightly more susceptible to structural failure.Within both structural types, the influence of design experience on the structural failure is limited. Furthermore, the effect of normal supervision on the failure probability in comparison to a process with only self-checking is about a factor 2,4 to 4,0. A design process without supervision and self-checking results in an unrealistic failure probability. However the occurrence of this seems not logical as self-checking is always present, mostly in a subconscious manner.