Research On Block-Structured Process Mining Techonology For Business Process Modeling

With the development of market economy, enterprises or institutions need to adapt to changes of market demand and the quick adjustment of their business is needed in order to improve rapid response capabilities of them. So, some parts in enterprise information systems supporting the business running are indeed to be reengineered or optimized. Nowadays, the refinement of business processes running mode was depended on the cooperation both business personals and technical personals. The results of the cooperation are very subjective and the lack of an overall, perfection. But we know that enterprise information systems with or without an explicit process almost provide the function to record alternation among tasks about contemporary enterprise business running. These records are also called as process logs. How to deduce the process model applying in process reengineering or process flexibility management through analyzing the process logs is an urgent need to address the problem. Business process models extracted from the process logs can be a true reflection of the implementation of enterprise business processes to support rapid business process modeling.The foundation of our work is simple format process logs that are not merely limited to or come from WFMS. Nowadays many enterprises or organizations have their own information systems using transactional systems such as ERP, CRM, B2B and SCM etc. These systems almost provide the simple format log information that our work needs. That is only to collect information of record alternation among tasks, i.e. process logs. Process logs record the reality of the implementation of business process. The business process model getting from process logs is more credible and easier to read.Process model often includes the complex structure, such as loop, parallel, alternative and non-free choice etc. The existing literature mining methods on these block structures are not well. There is no universal approach to solve the problems. In addition, the time information can be utilized to find out more precise logical relations of activities and calculate the performance of the process running.The work of our research is:deep study and refine on the process mining concepts of process mining technology, mathematical representation of process logs, Petri net representation of process model and the concepts of structural workflow net. Relational mathematical representations and definitions are presented. The main contributions of this thesis are as follows:1. At present, there is no uniform log format specification standard of the enterprise applications software. There are two kinds about log format:untimed and timed and both have noise in original logs. A block-structured mining approach (BRC) based on untimed from simple format logs of process was proposed for mining the relations of tasks in order to build business process model.The basic block structures are four kinds:sequence, parallel, alternative and loop. The approach first defined what kinds of process can be rediscovered, i.e. structural workflow (SWF). The basic idea of this approach is through the analysis of four kinds of structures existed in process; we can get structure modules and replace the log with the modules. Then, we begin the work of loop by using four mining algorithms to mine basic block structures from process log. At last, we get a minimal log until new basic block structure from it can not be found. After all work had been done, we use the modules to replace the tasks nodes. Finally, an understandable workflow net will be gotten. In the end, we prove that a reasonable and secure model can be gotten by using our mining approach. During the mining process of sequence or alternative tasks, thinking about the noise coming from other parallel tasks, the equivalence classes of parallel tasks were be proposed and used in BRC. First, it is done to part the parallel tasks to create equivalence classes of parallel tasks. By searching every equivalence classes, sequence or alternative relations can be mined from these equivalence classes. That is effective to solve the disturbance problem of parallel tasks during the block structure mining. Through the generating and running a certain number of experimental data, the result about mining quality inspection analysis show that BRC approach has obvious advantages in the mining of complex process. The causal relation mining with noise, parallel relation mining with noise, alternative structure mining with noise were discussed. And improved BRC algorithm with noise was proposed.2. Mining parallel relation in untimed logs is difficult and relatively large amount of calculation. An interval-based mining approach based on timed from simple format logs of process was proposed for mining the relations of tasks in order to build business process model.The idea is unlike the event based models. In the event based models, the task execution as inseparable, is an atom. But based on the interval of logs mining approach, each task execution has an interval based on its starting and ending events that are present in the process logs. This model can expand to be more accurate model. Among expanded model, the activity is a time intervals along the time axis, and two intervals can cross. The interval method is divided into two steps:(1)generate process execution graph. A DAG graph was produced for every execution. The implementation of the single task that has crossover makes the mining parallel relations easy. We combine execution graphs that have the same set of activities to construct a single DAG. (2)merger process execution graph. For a sub graph of process model, the execution of the various operations on it covered the same activity set. Namely business process using different parameters to choose partners has the same edge. Therefore the DAG produced by the combining is the control flow graph of process. Finally, through the merger DAGs depending on log, the process model graph is produced.The two kinds of algorithms are compared:interval algorithm and non-interval algorithm. Some meaningful conclusions are shown, such as the rate of both missing edges and excess edges and the size of log, the rate of both excess edges and the size of log, etc.3. Generally, logs have noise and were uncompleted. A method of mining logic relations of tasks from noise logs was given through discussing noise logs. Define a concept:sequence relation. This relationship based on two kinds of events:START and COMPLETE, and their time stamp. Then the measurement of sequence relationship is given. A mining method of process model graph is proposed. The original process model includes all kinds of the basic structures and the non-free choice.The method uses the process log having timestamp of task state, e.g. the average waiting time, the average execution time. During mining, average table (AVT) is constructed for each task by scanning a log firstly. Based on AVT, it is divided into two steps:(1) mining basic structures, e.g. sequence, parallel, alternative and iterative.(2) analysis the advance structure, e.g. OR-split. Through the above two step, the process model is expected. In addition, use of mining algorithm, the noise problems are solved well through the observation initial value. A comparison of the proposed algorithm and a-algorithm is done. TWM can mine the all processes that a-algorithm can mine, however a-algorithm fails to mine the all processes that TWM can mine. TWM and a-algorithm has an asymptotically equivalent complexity. The proposed algorithm provides useful information to solve hidden tasks.In addition, a process mining system supporting process mining algorithm is designed and implemented. The main purpose of the system design is:(1) verify the effectiveness of the a-algorithm; (2) comparing with our algorithm for getting effective reliability; (3) for complex variable structure process verify the limitations of the a-algorithm; (4) accumulating corresponding experience for process mining commercialization.The purpose of our work is to explore process mining technology supporting business process modeling, hoping to provide any valid way of solving the field's problems.