2nd International Workshop in Geoenvironment and Geotechnics, September 2008, Milos island, Greece

Artificial intelligence in longwall support effectiveness assessment

I. Andras
University of Petrosani, Romania

Gy. Deak
EcoMining Platform, Romania

V. Iordachita
ECOMITECH, Romania

ABSTRACT The paper deals with the application of artificial intelligence methods in assessing the effectiveness of longwall shield support. A fuzzy sets based method is used to derive response surfaces/curves representing the dependence between basic parameters of the shield, as the initial setting pressure and the stiffness on the response metric, i.e. the convergence and load increase during a working cycle. 1. INTRODUCTION The adaptation of powered roof support, from a constructive and functional point of view to the variable and specific of geologic and mining conditions, is a very important research subject. In past decades, the coal extraction technology has evolved dramatically. However, the problem of the interdependence between geo/mining conditions and the technical characteristics and functions of powered support represents a challenge to the specialists. To solve the related problems and provide manufacturers, designers and users scientifically sound solutions, one can employ new tools developed by engineers for an interdisciplinary approach. It is well-accepted that it is difficult to obtain closed form solutions from deterministic models. Also, historical statistical data represent a wide variability, so it is very difficult to describe the behavior of the rockmass and support system using a classical approach. In the present paper we try to use fuzzy modeling to obtain some qualitative results. 2. FUZZY MODELS FOR ROOF SUPPORT BEHAVIOR On one hand, operating characteristics of support systems are given in ranges. On the other hand it is difficult to quantify the parameters describing the geo mining conditions; an approximation is usually expressed by qualitative attributes. Hence, the decision to select a shield or to test the compliance of a given shield to the working conditions and technological factor can be made using fuzzy rules. Starting from the idea presented in Barczak (2006), we have developed an IT system based on fuzzy logic using the fuzzy module of MATLAB, using the concept of ground response curves. This system allows the establishment of the main parameters, i.e. the resistance and the stiffness and also allows setting the pressure for an appropriate selection of the shield. In Figure 1, the ground response curves for

Figure 1: Conceptual model for ground response curves.

2nd International Workshop in Geoenvironment and Geotechnics, September 2008, Milos island, Greece

supports with the four combinations of stiffness and yield loads, with roofs of different stability are depicted. Curves 1 to 4 represent the dependence between the roof convergence and the support load, for decreasing roof stability. The shape of the curves is determined by the empirical observation stating that at constant support load the convergence increases, when stability decreases and to maintain a given allowable convergence the support load must be higher for decreasing stability. The slope lines continued by horizontal lines represent the supports loading characteristic, and as the stiffness increases the line becomes more vertical. The elevation of the horizontal segment represents the value of the yield load of the support. The setting load is represented by the start point of the slope line on the vertical axis. The intersection between the support characteristic line and the roof characteristic curve gives the functioning point of the support-roof system at the equilibrium. The target for a proper support of the roof is to maintain this point on the inclined line segment. For this reason the external control parameters are the setting load, the yield load and the stiffness of the support shield. The main finding of previous research is the fact that the characteristic curve of the roof is changing in time during a working cycle, and the combination of the above mentioned parameters of the support must be selected in such a way, that the convergence be maintained under an imposed limit and the stability preserved during the entire cycle. The roof stability, described by (Esterhuizen and Barczak, 2006) is another metric which can be used as output for the devised fuzzy model. For illustration, this concept of stability in ground response curves is presented in Figure 2. In this approach, the curves represent the load-convergence dependence of the whole support- roof system. Different curves represent the systems behavior in different operating stages of the face. Between these three input parameters, i.e. the setting pressure (resistance), the yield pressure (resistance) and stiffness and the output parameters, i.e. stability and convergence, the field observations and the above common sense findings allow to derive inferences for fuzzy rules.

Figure 2: Conceptual model for ground response curves (Barczak, 2006).

Based on the above considerations, we developed two fuzzy models. The fuzzy models have been developed using MATLABs fuzzy toolbox. In the first model developed, we used inference rules for deriving the support loadconvergence curve and, respectively, the roof load-convergence curve. The output graphs are presented in Figures 3 and 4. In the second more sophisticated model, we used stiffness, stability, yield and setting load as the input variables and convergence as the output variable. The authors obtained the spatial graphs presented in Figures 5 to 7. Interpreting the results starting from these spatial graphs may offer some practical rules about prior selection of supports, using statements from historical data and simple factual reasoning.

Figure 3: Load-convergence curve of the support.

2nd International Workshop in Geoenvironment and Geotechnics, September 2008, Milos island, Greece

Figure 4: Load convergence curve of the roof. Figure 7: Setting load as a function of stability and stiffness.

Figure 5: Setting load as a function of stiffness and yield load.

It is possible to adjust and refine the model, comparing field data with those obtained from the model, with crisp values if applicable. The authors can use the defuzzyfication module of the model as an interactive tool to simulate different situations by modifying some

Figure 8: Defuzzification module as an interactive tool.

input parameters and derive specific values for outputs, as in Figure 8. 3. CONCLUSIONS The adaptation of powered roof support, from design and operational point of view to the variation and specificity of geologic mining conditions, is a very important research subject. The use of closed form equations, analytical models, or static approaches is limited and sometimes useless for practical applications. It is possible, using the findings from experience, conceptual and qualitative models to develop simple recommendations for the preliminary selection of the proper support features, and roof characteristics described by non specific variables using fuzzy reasoning. The results obtained with this first attempt to use fuzzy sets to model the longwall supports are encouraging to continue the research.

Figure 6: Setting load as a function of stability and yield load.

2nd International Workshop in Geoenvironment and Geotechnics, September 2008, Milos island, Greece

The set of rules used in fuzzy inference can be eventually derived by employing the neural net method, using recorded time series of the main parameters of longwall supports. REFERENCES
Esterhuizen, G.S. and T.M. Barczak, 2006. Development of Ground Response Curves for Longwall Tailgate Support Design, Proceedings of the 41st U.S. Rock Mechanics Symposium, Golden, Colorado, June 1721, 2006. Alexandria, VA: American Rock Mechanics Association, 2006 Jun; 1-10. Barczak, T.M., 2006. A Retrospective Assessment of Longwall Roof Support with a Focus on Challenging Accepted Roof Support Concepts and Design Premises. 25th International Conference on Ground Control in Mining, 232, 244.