Please take a look at the NNAPICS website also.
PROJECT SUMMARY
The presence of impurities can have damaging effects on the quality of products made with Portland cement, which are difficult to predict. Nevertheless, introduction of impurities into cement systems is inherent to recycling of industrial by-products by utilisation in cement-based building materials and in treatment of industrial wastes by cement-based solidification prior to disposal. The consequences of design of cement-based products without proper consideration of the potential for complex interactions between cementing components and impurities are: handling difficulties, failure to set, improper strength development, and deterioration over time. These hazards, and the difficulty in predicting their occurrence, have hindered both utilisation and/or solidification of industrial by-products, because potential benefits are outweighed by the expense of failures.
The objective of the 41-month NNAPICS project (12/1997-4/2001) was to collect and use existing data from the literature and supplementary data from a laboratory programme to examine the application of neural network analysis for predicting interactions in, and final properties of, cement-based products containing impurities. The projected was conducted by a consortium of 8 partners, under the European Commission's Industrial and Materials Technologies Programme.
The main achievements of the project were:
1) creation of the MONOLITH database of 1506 literature references and properties of 7953 cement-based products containing impurities, which represents a large proportion of the information available in the literature,
2) development of the MONOLITH interface, which allows flexible search, output and viewing of the data in the database, as well as providing example neural network models, and enabling storage of new data by future users,
3) the findings of neural network analysis of cohesive data subsets extracted from the MONOLITH database, and
4) the findings from the laboratory programme, for 230 cement/waste products.
The NNAPICS project final report, and the MONOLITH database and user-interface, can be obtained at no charge by registering with the project coordinator:
We will automatically register you as a member of the NNAPICS End-User Group. If you do not wish to be a member, please mention this in your e-mail. We will send you the desired information regardless.
The information is available on this CD-rom. To use it you have to register -at no costs- with the project coordinator. Please click here for the permission to use the database. The database can be found in the MONOLITH2001.1 folder, the end-report in the NNAPICS REPORT folder on this CD-rom. Please note that if you want to use the database you must have Microsoft® Access 2000 installed (lower versions than 2000 will not work with the database).
The findings from neural network analysis and the laboratory programme have been published in the open literature (see Publications).
Examination of the literature data in the MONOLITH database showed that researchers measure an extremely wide variety of properties for an equally wide variety of materials. The properties and materials for neural network analysis were chosen based on their practical importance and availability in the database. Neural network models were constructed for prediction of:
1) setting time of calcium aluminate cements containing contaminants,
2) unconfined compressive strength (UCS) of Portland cement containing contaminants,
3) leachate pH for Portland cement containing contaminants, and
4) UCS of Portland cement containing electric arc furnace dust, foundry dust, municipal waste incinerator fly ash, other ashes and plating sludge.
The effects of each of the input variables on the model predictions was examined. It was found that construction of successful models was possible, with prediction errors approaching experimental error, and that modelling was useful for generalising about the relative effects of the input variables on the outputs using the results from different studies. The work has shown that the potential for practical implementation of models of this type in prediction of long-term durability, and/or formulation design in waste treatment facilities clearly exists, but more detailed definition of the dataspace by experimentation, with more complete harmonisation of methods and measurement and reporting of experimental variables, will be necessary before reliable, trustworthy models can be developed. In the meantime, neural network models are useful as a research tool, which can highlight important variables for design of formulations and test methods, and guide experiments to investigate interference and contaminant immobilisation mechanisms. Also, the MONOLITH database and interface provides a building block that can be helpful to both researchers and industry in advancing this field.
The NNAPICS consortium partners are interested in possible further collaborations in this area. If you have an idea to work together, please contact the project coordinator at julia.stegemann@eng.ox.ac.uk.