Neural networks in process life cycle profit modelling Neural networks in process life cycle profit modelling Teemu Räsänen a,*, Risto Soukka b, Sami Kokki b, Yrjö Hiltunen a a Department of Environmental Science, University of Kuopio, P.O. Box 1627, FIN-70211, Kuopio, Finland b Department of Energy and Environmental Technology, Lappeenranta University of Technology, P.O. Box 20, FIN-53851, Lappeenranta, Finland Abstract Changes in operational environment of the process industry such as decreasing selling prices, increased competition between compa- nies and new legislation, set requirements for performance and effectiveness of the industrial production lines and processes. For the basis of this study, a life cycle profit (LCP) model of a pulp process was constructed using different kind of process information including chemical consumptions and production levels of material and energy flows in unit processes. However, all the information needed in the creation of relevant LCP model was not directly provided by information systems of the plant. In this study, neural networks was used to model pulp bleaching process and fill out missing information and furthermore to create estimators for the alkaline chemical consumption. A data-based modelling approach was applied using an example, where factors affecting the sodium hydroxide consump- tion in the bleaching stage were solved. The results showed that raw process data can be refined into new valuable information using computational methods and moreover to improve the accuracy of life cycle profit models. � 2007 Elsevier Ltd. All rights reserved. Keywords: Life cycle profit modelling; Neural network; Multi-layer perceptron; Variable selection; Pulp industry 1. Introduction In the process industry the development of competitive- ness of an existing production line has become a lifeline for many production units, because competition between companies has increased, new legislation has been pub- lished and the consequences of disturbances are bigger than before. Thus, it is essential to find process developing possibilities to increase profits and maximize returns and at the same time eliminate risks. These new demands set requirements not only for performance and effectiveness of processes but also for optimizing process life cycle profits. However, the construction of large economical models of processes is too complicated using only process models which are based on natural laws. The connections between the units of the processes in large scale plants are usually so complex that modelling based on natural laws is unable to produce all the needed information for a decision-making process. On the other hand, modelling on the basis of nat- ural laws requires lots of resources, like time and staff resources as well as economic and computational resources, and due to the these reasons modelling on sche- dule is often not possible (Soukka, 2007). The applicability of life cycle modelling principles to the modelling of increasingly complex processes has been examined. Life cycle modelling (life cycle assessment and life cycle costing) collects data on the entire life cycle. For this reason it is obvious that everything cannot be modelled on the basis of natural laws. Life cycle modelling requires the easy collection of an immense amount of infor- mation on the different stages of the process life cycle, so that it is impossible to produce vast amounts of measure- ment data due to the restrictions of the resource reserved for the report. Thus life cycle assessment also uses esti- mated, calculated and literature-based information (Sou- kka, 2007). 0957-4174/$ - see front matter � 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2007.07.006 * Corresponding author. Tel.: +358 44 7162337; fax: +358 17 163191. E-mail addresses: teemu.rasanen@uku.fi (T. Räsänen), risto.soukka@ lut.fi (R. Soukka), sami.kokki@lut.fi (S. Kokki), yrjo.hiltunen@uku.fi (Y. Hiltunen). www.elsevier.com/locate/eswa Available online at www.sciencedirect.com Expert Systems with Applications 35 (2008) 604–610 Expert Systems with Applications mailto:teemu.rasanen@uku.fi mailto:risto.soukka@ lut.fi mailto:risto.soukka@ lut.fi mailto:sami.kokki@lut.fi mailto:yrjo.hiltunen@uku.fi The life cycle profit (LCP) analysis has been developed for the purpose of evaluating the economic impacts of process changes with a model based on the production principles of information applicable in life cycle modelling. A life cycle profit model is based on several kinds of infor- mation. The measurement data of one point can be used to constitute a statistic distribution. The estimated informa- tion and data from literature are in turn information that can be regarded as unique (Soukka, 2007). Among the other things, data-based computational methods can be used in the production of calculated information and, par- ticularly, in the identification of consequence-related effects including solving material and energy flows of the process (Räsänen et al., 2006). It is essential that life cycle profit model has been con- structed so that it reveals how changes are affecting reve- nues and costs of a production line. On the one hand, life cycle revenues can be increased by removing bottlenecks and production failures. These problems contain a high risk to cause production losses. On the other hand, costs can be reduced for example by affecting energy, chemical or water consumption of processes and the amount of efflu- ents from processes. Raw process data is one of the LCP models information sources and it can be used various ways. Industrial pro- cesses produce typically a huge amount of measured data describing each process performance. Several studies (Hei- kkinen, Kettunen, Niemitalo, Kuivalainen, & Hiltunen, 2005; Hiltunen et al., 2006) have shown that data-driven approaches are a fruitful way of developing for example the process state monitoring. Archived process data is also an important resource for the knowledge management of the process and it can be used for the optimization and improvement of productivity. In this study, we combine data-driven modelling and process life-cycle profit modelling into more intelligent sys- tem to produce valuable and more accurate new informa- tion to decision making. The proposed methods were applied and tested using the data set of an industrial scale pulp bleaching process. 2. The process and the data In this study life-cycle profit model (Soukka, 2007) was constructed for a paper mill, which produces newsprint and directory paper, coated and uncoated fine paper, core- board and spruce timber. The paper mill contains also pulp production and refined mechanical and ground wood pulp are produced for newsprint paper production. Also ECF- pulp is produced for the fine paper production. The life cycle profit model covers the whole production line of the pulp. However, in this study we concentrated on solving the missing or unknown correlations between pro- cess variables in the pulp bleaching process which is shortly described in Fig. 1. The objective of bleaching is to improve the brightness and the cleanliness of the pulp. The raw process data was originated from measuring systems and it was collected from pulp mill databases. The used data set contained 35 variables and 1500 rows. The data set consists of the time period where pine was used as pulp raw material and the time resolution was one hour. Variables were selected by process experts. 3. Methods 3.1. Life-cycle profit modelling The main purpose of the plants life cycle profit (LCP) model is to recognize the development possibilities, which can be achieved by process changes allocated to different stages of the process. The LCP model shows how life cycle profit varies in consequence of the process changes and owing to this it is possible to prioritize different possibilities to develop the process. The LCP analysis is originally used for modelling existing production lines but the model can also be used for strategic planning of new production lines. Therefore, it is possible to evaluate risks that can be actu- alized if big changes in the proportion of prices happen. Hunkeler (1999) has been presented that Life Cycle Profit- ability is an evolutionary means to conduct business prac- tice under scenarios where envirotechnical imperatives Fig. 1. Pulp bleaching stage contains several stages because target brightness of the end product cannot be achieved in only one bleaching step. Pulp is also washed between each stage. T. Räsänen et al. / Expert Systems with Applications 35 (2008) 604–610 605 https://isiarticles.com/article/1515