Case-based Reasoning for Separation Process Synthesis

A method of reusing existing design cases for synthesis of distillation process sequences is presented. The approach uses case-based reasoning (CBR) that finds the most similar existing separation designs and applies the knowledge of their concept for solving new problems. The method has previously been developed for selection of single separations and simple sequences but has now been extended to cover synthesis of more complicated sequences. The method is intended for finding feasible process alternatives in preliminary process design. INTRODUCTION The paper presents a method for finding feasible separation process sequences by using case-based reasoning (CBR). CBR is a method of reusing existing design cases for making new designs. This means finding most alike existing processes and applying the knowledge of their separation capacity and design for solving new design problems. This is especially important in the early phases of process design when many alternatives should be quickly screened before a more detailed study is done. There is a great need for these kinds of screening tools to reduce the number of design options and quicken the process design in practice [1]. When dealing with multicomponent mixtures, the number of possible separation methods, their combinations and process structures to be screened is huge as well as the work involved. The problem of synthesis of a simple separation sequences and the selection of single separations has been studied earlier [2, 3, 4]. However a comprehensive CBR-based methodology for the synthesis of more complicated separations sequences has not been presented yet. CBR IN PROCESS SYNTHESIS Case-based reasoning (CBR) is one of the non-symbolic AI methods [5]. CBR solves new problems by finding and adapting existing successful designs for solving new problems (Fig.1). The lack of systematic reuse of existing design experience has been a shortage in process design. The main benefit of CBR approach is that readily available existing knowledge can be utilised systematically also in large and complex problems such as process synthesis and design. In this way the time-consuming conceptual screening phase of a design project can be fastened. CBR is attracting attention, because it seems to directly address the conceptual process design problems outlined earlier. Some benefits of the CBR approach are: 1. CBR does not require an explicit domain model and so elicitation becomes a task of gathering case histories. 2. Implementation is reduced to identifying significant features that describe a case, which is an easier task than creating an explicit model. 3. By applying database techniques, large volumes of information can be managed. 4. CBR systems can learn by acquiring new knowledge as cases, thus making maintenance easier. RETAIN Another design to adapt Problem RETRIEVE Case-Base consists of old designs Proposed solution Confirmed solution Could be saved as a case. REVISE The user checks the solution Figure 1. Principle of case-based reasoning Because generalisations are not needed in CBR, no data is lost. CBR gives answers to design problems in a straightforward way. The results are dependent on the retrieval parameters and the adaptation applied. The strong interaction with the user makes the flexible and interactive use of existing data and design experience possible. The CBR search can be focused on different aspects by defining new search criteria and weighting retrieval criteria differently. In this way the same case base can be used for several types of tasks. The system learns by updating the information of the database. CASE RETRIEVAL A retrieval algorithm using the indices in the case-memory should retrieve the most similar cases to the current problem or situation. The retrieval algorithm relies on the indices and the organisation of the memory to direct the search to potentially useful cases. The issue of choosing the best matching case has been addressed by research on analogy. This approach involves using heuristics to constrain and direct the search [5]. Methods for case retrieval are nearest neighbour, induction, and knowledge-guided induction and template retrieval. These methods can be used alone or combined into hybrid retrieval strategies. If the nearest neighbour is used, then case features should be able to be weighted and similarity measures customised. If inductive techniques are used, the index tree generated should be open to inspection and alteration by developers. A typical equation for calculating nearest neighbour matching is Eq. 1, where w is the importance weighting of a feature, sim is the similarity function, I f and R f are the values for feature i in the input and retrieved cases respectively [5].