1 GOSPODARKA SUROWCAMI MINERALNYMI Tom Zeszyt 4/3 R. ZIMROZ* Adaptive approaches for condition monitoring of mining machines Introduction Mining machines often work under non-stationary, Time-Varying Load Conditions (TVLC). The non-stationarity of load is mainly caused by the non-stationarity of technological process of mining (variation of external load caused by an operating bucket wheel, a time-varying stream of materials transported by conveyors, etc.) (Bartelmus, Zimroz 2006a, b, c). Taking into account variation of external load, complex mechanical structure of diagnosed object, enviromental impact, high level of internal/external interference, possibility of multi-faults appearance (in different location, development stage, nature of damage, ie distributed/localised, etc) condition monitoring becomes a serious task. Indeed, diagnostics of wear and especially of local damages (for example related to tooth crack/ /breakage) is very complicated and unfortunately even proffesional diagnostics systems are sometimes not able to recognize it. In author s opinion one of possible reasons is that systems are closed, with rigid rules of signal processing and reasoning. For example for such unique machines like bucket wheel excavator there is a need for more individual approaches. In this paper an attempt of generalization of requirements for such diagnostic system is presented. First assumption is that condition monitoring should be adaptive at signal preprocessing level, diagnostic feature extraction level and finally at diagnostic reasoning level. As practical results of research carried out in this field following issues be presented: adaptive signal preprocessor (adaptive filter) for local damage detection in gearboxes and bearings (G, B), adaptive subsystem for local damage detection (G, B), adaptive subsystem of diagnostic reasoning for local damage diagnostics (G), * Department of Mining, Wroclaw University of Technology, Wroc³aw Poland.
2 104 adaptive subsystem of diagnostic reasoning for geared wheel cooperation assesment (G). It will be illustrated by examples based on industrial data captured from mining machines during normal (non-stationary) operation, namely two stage gearbox used in driving unit for belt conveyor transportation, complex gearbox with damaged planetary stage (bucket wheel excavator), bearing used in pulleys (belt conveors). 1. Identification of operating condition In this section methods of operating condition identification will be presented. Depends on machine type, rotational speeds, nature of external load variation and technical problems related to signal acquisition different approaches will be presented. As indicators of operating condition one can use as follows: current consuming by motor or input rotational speed (requires additional channels) or information extracted from vibration signal (load variation causes modification of vibration signal properties) (Stander 2002; Bartelmus, Zimroz 2006b; Zimroz 2007a). Figures 1, 2 show examples of current (consumed by motors) variation, speed profiles (variation of input speed due to non-stationary operation) and information about load variation extracted from vibration signal. Operating condition may vary in wide range depends on machine type and some factors related to technological process, for example, for bucket wheel excavator (BWE) load variation depends on excavated material properties and the way of digging process, for belt conveyor systems (BCS) external load of driving unit depends on actual output, environmental issues etc (Fig. 3) LOAD (current consumed by motor) bigger Current [A] Z11: current [A]- motor A Z11: current [A]- motor B Z11: current [A]- motor C : : : : : : : : : : :14-20 time Fig. 1 Example of current signals for a) bucket wheel excavator, b) 3 motors on driving station for belt conveyor system Rys. 1. Przyk³ad sygna³ów pr¹dowych dla a) koparki ko³owej, b) 3 silników na stacji napêdu systemu przenoœnika taœmowego
3 105 Fig. 2. Example of a) speed profile for bucket wheel excavator, b) load variation extracted from vibration signal for bucket wheel excavator Rys. 2. Przyk³ad a) profilu prêdkoœci dla koparki ko³owej, b) zmiennoœci obci¹ eñ pobranych z sygna³u wibracji dla koparki ko³owej Fig. 3. View of a bucket wheel and a belt conveyor during operation Rys. 3. Widok ko³a maszyny i przenoœnika taœmowego podczas pracy Operating condition for BWE may reach even critical value that stops operation (it happens in case of difficult-to-mine materials or if bucket meets a stone in excavated material) Bucket wheel excavator Identification of operating conditions for BWE may be related to cyclic nature of digging process. Auxilary signal (current, speed profile) is divided into small segments, for each segment simple statistics are calculated (mean, standard deviation). A method proposed by Stander (Stander 2002) (extraction of information from vibration signal) is useless due to specific, complex design and low speed rotating components signal cannot be separated due to interferences.
4 Belt conveyor For conveyor systems there is no cycle, nature of operating conditions variation is random and slowly changeable. In such situation segmentation should be done by analysis of data properties. In this case current profile or speed profile was not available. For identification of TVLC one may find that information about instantaneous speed may be obtained by vibration signal analysis without tacho signal, as it was shown in Bonnardot work (Bonnardot et al 2005). In latest work a simple approach was used that assumes acceptable level of speed vatiation during 5s segment. Stationarity of instantenous speed in segment of signal means that it is possible to recalculate it from mesh frequencies detected in Fourier spectrum of signal. 2. Adaptive approach for local damage detection in mining machines Local damages detection in gearboxes or bearings is imporant task in condition monitoring. Especially damage at early stage produces low energy, non-linear phenomena that may be dificult to identify based on raw vibration. Advanced signal processing should be used in such situation one of them is adaptive filtering (AF) that can separate signals based on their statistical properties (deterministic or random) (Widrow 1985; Haykin 1996). In next section application of adaptive filtering for bearing damage detection is presented, after that, the same approach is used for planetary gearbox diagnostics. Scheme of adaptive filtering system is presented in Figure 4. Fig. 4. Scheme of adaptve filtering system Rys. 4. Schemat adaptacyjnego systemu filtruj¹cego 2.1. Signal extraction by adaptive filtering of raw vibration (bearing from pulley) Local damage detection for bearing used in pulleys is dificult due to high level of interference in captured signal. Source of mentioned interference is a gearbox vibration with
5 107 Rys. 5. Original signal and extracted signal by adaptive filter (ZMIEN OPIS) Rys. 5. Sygna³ pierwotny i wydobyty przez filtr adaptacyjny (ZMIEN OPIS) high amplitudes of components related to mesh (fundamental and harmonics) frequencies. Signal of interest (contains information about damage) has lower energy than interference and his nature is random (in oposite to detrerministic gearbox signal). Similar problem was formulated by Antoni and Randall (Antoni, Randall 2002, 2003) for helicopters. On Fig. 5 results of signal extraction based on AF is presented. Clear impact visible on Fig. 5 (right) are related to local damage in bearing Signal extraction by adaptive filtering of raw vibration (planetary gearbox) In this section diagnostic task is to extract impulsive signal related to local damage from complex vibration signal produced by multistage planetary gearbox. Diagnostic task is similar to previous: signal structure (in frequency domain) is very rich and there is a need to extract random low-energy, cyclic and wideband excitation from signal significantly corrupted by other deterministic sources (mainly by signal related to mesh frequencies of planetary stage). The difference between amplitudes of raw and extracted signal is 10 times (Fig. 6). Similar issue was considered in (Lee, White 1998) Fig. 6. Original signal and extracted signal by adaptive filter Rys. 6. Sygna³ pierwotny i wydobyty przez filtr adaptacyjny
6 Adaptive subsystem for local damage detection (two stage gearbox) As it was said, properties of diagnostic signal depend on damage size (it is basis for diagnostics) and also operating condition. Let s consider two stage gearbox used in drivining unit for belt conveyor. For relatively stationary operating conditions there is a need to find optimal filter that will be able to extract signal of interest with as high as possible value of kurtosis (statistical parameter ussually used as an indicator of spikeness in vibration signals) and/or value of sum of components (sidebands) in envelope spectum (Fig. 7). Details of this Fig. 7. Original signal and extracted signal by optimized stationary filter Rys. 7. Sygna³ pierwotny i wydobyty przez zoptymalizowany filtr stacjonarny Fig. 8. Adaptive system for local damage detection for gearboxes used in conveyors Rys. 8. System adaptacyjny dla wykrywania uszkodzeñ lokalnych skrzynek przek³adniowych stosowanych w przenoœnikach
7 109 approach, criteria selection and more practical consideration may be found in (Bartelmus, Zimroz 2006d; Zimroz 2007b) For nonstationary operation it may happen that properties of signal will change significantly. Research carried out by (Bartelmus, Zimroz 2008b) shows that so called optimal frequecy band (OFB) i.e. frequency band that contains information about local damage is very sensitive to external load value. Advanced concept of searching OFB for slowly varying external load is presented on Figure 8. First part of system is trying to extract signal of interest by filtering of raw signal at OFB, next extract values of features and finally decision making subsystem is providing diagnostics decision. It works very well for stationary load. For time varying operating condition it may appear that OFB will be not optimal anymore. Adaptive approach in this case is usage of second part of system for searching variation of OFB. If external load has change, OFB has change so second part of system will find it and tune adaptive filter in first part of system. It is obvieus that variation of load has to be slow, that is the case of conveoyr system (for BWE not). 3. Adaptive approach for distributed damage diagnosis in mining machines 3.1. Bucket wheel excavator As distinc from local damage diagnostics, task for distributed faults is not only detect but also estimate size of change of condition (for example wear). Unfortunately diagnostic features ussually amplitudes of specific components in power spectral density (PSD) of raw signal (components related to diagnosed rotating part of machine) are very sensitive to load value (Bartelmus et al. 2007; Bartelmus, Zimroz 2007). Moreover, variation of operating condition often appears as simultanous load and speed variation, so amplitudes in PSD are different due to load variation and also due to smearing (frequency modulation) caused by speed variation. If smearing effect is present so called order analysis (instead of classic frequency analysis) should be used (Gade 1999). Figure 9 shows a concept of adaptive system for bucket wheel excavator. Operating condition is identified by input instantenous speed monitoring (based on so called tacho signal serie of rectangular pulses related to input shaft revolution). System exploits idea of segmentation of signals; segments are related to digging cycle. For auxiliary signal speed profile is obtained, then, for each segment, an averaged value of speed and standard deviation of speed are calculated. For vibration signal classic spectrum is used as a basis feature extractor. It was experimentally proved that for short segments smearing effect can be neglected (Fig. 10). Segments for no-load and small load cases are deleted (signal verification) due to unsatisfied damaged-undamaged separation ability.
8 110 Fig. 9. Adaptive system for gears diagnostics for nonstationary operations Rys. 9. System adaptacyjny dla diagnostyki przek³adni dla pracy niestacjonarnej Fig. 10. Examples of spectra for segments (bad condition) smearing neglectible for short segments Rys. 10. Przyk³ady widma dla segmentów (z³y stan) rozmazania pomijalne dla segmentów krótkich 4. Adaptive diagnostic reasoning techniques In previous sections a set of techniques and methods for signal preprocesing and feature extraction was presented. For clasic condition monitoring, diagnostic decisions are based on simple comparison of feature value with threshold (s). As it was underlined mining machines work under nonstationary operation and diagnostic reasoning rules must take into account also value of operating conditions. In some cases there is a need to consider vector of features and vector of indicator of operation (advanced operating condition parametrisation). Diagnostic reasoning becomes multidimensional and complicated (Cempel, Tabaszewski 2007). A neural network, a fuzzy systems or generally data mining approaches are often used (Bartelmus et al. 2003; Bartelmus, Zimroz 2004). It is obievus that all intelligent approaches are adaptive (due to training phase). In this paper simple adaptive reasoning rules are briefly described.
9 D i s t r i b u t e d f a u l t s In the simplest case presented here, a distribution of features for good and bad condition as a function of input rotational speed is showed. As it was said for small load cases (speed > 995RPM) it is imposible to recgnize condition of machine. It has very physical explanation if consider dynamics of gear pair. For smaller value of speed (bigger load) separation is easier differences between amplitudes are bigger. By simple statistical data processing a linear approximation may be found for both conditions, after this speed dependent alarm level in the middle is obtainted. Adaptation in reasoning is that diagnostic rule depends on operating condition, i.e. different value of threshold is used for comparison with value of feature. Without adaptation for wide range of operating condition it is imposible to recognize correctly condition of machines. Fig. 11. a) DF-OC database for planetary stage, b) regression and OC-dependent alarm level Rys. 11. a) Baza danych DF-OC dla stopnia planetarnego, b) poziom alarmu regresji i zale nego od OC Similar experiments were carried out for gearboxes used in belt conveyor systems. Due to slow, non-cyclic variation of operating condition procedures of signals segmentation and operating condition identification were a bit different but idea of adaptive reasoning is exactly the same. As it was introduced by Bartelmus and Zimroz for planetary gearbox, intensity of feature-operating condition dependency may be used as a indicator of technical state of machine (Bartelmus, Zimroz 2008a). In other words if machine is sensitive to load variation (generates vibration depending of load) it means change of condition of this machine. It is clear that for both planetary and fixed axis gearbox from conveyor slopes of regression lines are significantly different.
10 112 Fig. 12. Diagnostics relation for belt conveyor system working under TVLC a) bad condition, b) good condition Rys. 12. Relacja diagnostyki dla systemu przenoœnika taœmowego pracuj¹cego w warunkach TVLC a) z³y stan, b) dobry stan 4.2. L o c a l f a u l t s In this section it will be shown that for local fault dependency between feature and operating condition also exists. Zimroz (Zimroz 2008) has analysed it using developed in (Bartelmus 2001) model of driving units for belt conveyor. For combination of different value of external load L and different size of local damage (simulated by change of stiffness for one of pair of teeth) diagnostic signal were simulated. Next, using signal filtering at Fig. 13. Diagnostic relation between value of kurtosis and external load obtained from simulations Rys. 13. Relacja diagnostyczna pomiêdzy wartoœci¹ kurtozy i zewnêtrznym obci¹ eniem uzyskanym w symulacjach
11 113 optimal frequency band kurtosis value was used as a diagnostic feature for each signal. Results of experiment have been ploted kurtosis vs load. For each serie of data regression have been calculated. From Figure 13 one may easily find that for small change of stiffness (small damage) dependency is week and linear (0.0031x 2 0x 2 ). If size of damage is growing up, relation becomes stronger and non-linear (0.026x 2, 0.20x 2, 0.35x 2 ). In this case a novel approach may be defined as follows: indicators of technical state are intensity of feature-operating condition dependency (like for wear) but also level of nonlinearity of this relation. Conclusion Condition monitoring of mining machines under non-stationary operation has been considered in this paper. It was stated that due to complex mechanical structure of diagnosed object (design factors), variation of external load (operational factors), high level of internal/ /external interference, possibility of multi-faults appearance (in different location, development stage, nature of damage, i.e. distributed/localised, etc. change of condition factors) and environmental impact, condition monitoring for mining machines should be adaptive. A several examples that confirm this statement has been provided in this paper, namely: adaptive signal pre-processor (adaptive filter) for local damage detection in gearboxes and bearing (G, B), adaptive subsystem for local damage detection (G, B), adaptive subsystem of diagnostic reasoning for local damage diagnostics (G), adaptive subsystem of diagnostic reasoning for geared wheel cooperation assessment (G). All considered examples come from analysis of industrial data captured from mining machines during normal (non-stationary) operation. Paper is financially supported by State Committee for Scientific Research in as research project REFERENCES  A n t o n i J., R a n d a l l R.B., 2002 Differential diagnosis of gear and bearing faults. ASME Journal of Vibration and Acoustics 124 (2),  A n t o n i J., R a n d a l l R.B., 2003 Unsupervised noise cancellation for vibration signals. Part I evaluation of adaptive algorithms, Mechanical Systems and Signal Processing,  B a r t e l m u s W Mathematical Modelling and Computer Simulations as an Aid to Gearbox Diagnostics. Mechanical Systems and Signal Processing vol. 15, nr 5, s  B a r t e l m u s W., Z i m r o z R., 2004 Application of self-organised network for supporting condition evaluation of gearboxes. Methods of artificial intelligence. [AI-METH 2004], Gliwice, November/ Eds.T.Burczyñski,W.Cholewa,W.Moczulski. Gliwice, AI-METH Series.  B a r t e l m u s W., Z i m r o z R., 2006a Bucket wheel load variability identification on vibration analysis. Proceedings of the Fifteenth International Symposium on Mine Planning and Equipment Selection, Torino, Italy, September [Vol. 1 / Ed. by M. Cardu].
12 114  B a r t e l m u s W., Z i m r o z R., 2006b Identyfikacja warunków eksploatacyjnych na potrzeby diagnostyki przek³adni planetarnej do napêdu ko³a czerpakowego. Diagnostyka nr 1, Warszawa, s  B a r t e l m u s W., Z i m r o z R., 2006c Influence of random varying load on vibration signal generated by planetary gearboxes driving bucket wheel in excavators. Proceedings of the 19th international congress Condition monitoring and diagnostic engineering management. (COMADEM 2006), Lulea, Sweden, June / Ed. by U. Kumar, A. Parida, Raj BKN Rao. Lulea : Lulea University Press, pp  B a r t e l m u s W., Z i m r o z R., 2006d Optymalny zakres czêstotliwoœci w procedurze demodulacji amplitudy w zastosowaniu do uszkodzeñ lokalnych. Diagnostyka no 1, s  B a r t e l m u s W., Z i m r o z R., 2007 Adaptive method of reasoning for planetary gearbox working under time varying operation. Conference on Machine Diagnostics, Wegierska Gorka (in Polish).  B a r t e l m u s W., Z i m r o z R., 2008a Vibration condition monitoring of planetary gearbox under random varying. Mechanical Systems and Signal Processing (accepted for publication, April 2008).  B a r t e l m u s W., Z i m r o z R., 2008b Wykrywanie uszkodzeñ lokalnych w uk³adach napêdowych maszyn górniczych wybrane zagadnienia. Conference on Machine Diagnostics, Wêgierska Górka.  B a r t e l m u s W., Z i m r o z R., B a t r a H., 2003 Gearbox vibration signal pre-processing and input values choice for neural network training. Methods of artificial intelligence. AI-METH Symposium Proceedings, November 5 7 / Eds T. Burczyñski, W. Cholewa, W. Moczulski, Gliwice.  B a r t e l m u s W., Z i m r o z R., S a w i c k i W., 2007 Wp³yw zmiennych warunków eksploatacji na proces oceny stanu przek³adni planetarnych w uk³adach napêdowych ko³a czerpakowego koparek ko³owych. Diagnostyka no 1, pp  Bonnardot F., El Badaoui M., Randall R.B., Danieere J., Guillet F., 2005 Use of the acceleration signal of a gearbox in order to perform angular resampling (with limited speed fluctuation). Mechanical Systems and Signal Processing 19,  C e m p e l Cz., T a b a s z e w s k i M., 2007 Multidimensional condition monitoring of machines in non- -stationary operation. Mechanical Systems and Signal Processing 21,  Gade S., Herlufsen H., Konstantin-Hansen H., Wismer J., 1999 Order Tracking Analysis, B, K Technical Review No. 1.  H a y k i n S., 1996 Adaptive Filter Theory, Prentice-Hall, New Jersey.  L e e S.K., W h i t e P.R., 1998 The enhancement of impulsive noise and vibration signals for fault detection in rotating and reciprocating machinery. Journal of Sound and Vibration, 217 (3), pp  Stander C.J.,Heyns P.S.,Schoombie W.,2002 Using vibration monitoring for local fault detection on gears operating under fluctuating load conditions. Mechanical Systems and Signal Processing 16 (6),  W i d r o w B., S t e a r n s S., 1985 Adaptive Signal Processing, pp Englewood Clis NJ: Prentice- -Hall.  Z i m r o z R., 2007a Non-stationary operation condition analysis by instantaneous speed monitoring for mining machines diagnostics. Proceedings of the 16th International Symposium on Mine Planning and Equipment Selection, Bangkok, Thailand.  Z i m r o z R., 2007b Optymalizacja procedury demodulacji z wykorzystaniem unormowanej sumy amplitud wstêg bocznych jako kryterium decyzyjne. Górnictwo i geologia IX. Wroc³aw, Oficyna Wydaw. PWroc., s  Z i m r o z R., 2008 Diagnozowanie przek³adni zêbatych w uk³adach napêdowych przenoœników taœmowych Wykrywanie uszkodzeñ lokalnych w warunkach zmiennego obci¹ enia. Transport Przemyslowy 1 (31), pp
13 115 PODEJŒCIA ADAPTACYJNE W MONITOROWANIU STANU MASZYN GÓRNICZYCH S³owa kluczowe System adaptacyjny, kopalnia odkrywkowa, technika diagnostyczna Streszczenie Maszyny górnicze czêsto pracuj¹ w warunkach obci¹ eñ niestacjonarnych, zmiennych w czasie (TVLC). Niestacjonarnoœæ obci¹ enia jest spowodowana g³ównie niestacjonarnoœci¹ technologicznego procesu urabiania (zmiennoœæ zewnêtrznego obci¹ enia spowodowana przez pracuj¹ce ko³o maszyny, zmienny w czasie strumieñ materia³ów transportowanych przez przenoœniki itp.). Uwzglêdniaj¹c zmiennoœæ obci¹ enia zewnêtrznego, z³o- on¹ budowê mechaniczn¹ diagnozowanego obiektu, wp³yw na œrodowisko naturalne, wysoki poziom zak³óceñ wewnêtrznych / zewnêtrznych, mo liwoœæ wyst¹pienia uszkodzeñ wielokrotnych (w ró nych miejscach, rozwijaj¹cych siê, charakter uszkodzeñ, tzn. rozproszone / lokalne, itp.), monitorowanie stanu staje siê powa nym zadaniem. W istocie rzeczy, diagnostyka zu ycia, a zw³aszcza uszkodzenia lokalne (na przyk³ad dotycz¹ce pêkniêæ / z³amañ zêbów) jest skomplikowana i niestety nawet profesjonalne systemy diagnostyczne czasem nie s¹ w stanie jej skutecznie prowadziæ. Zdaniem autora, jednym z mo liwych powodów jest to, e systemy diagnostyczne s¹ zamkniête i maj¹ sztywne zasady przetwarzania sygna³ów i wyci¹gania wniosków. Dla przyk³adu, przy tak wyj¹tkowej maszynie, jak¹ jest koparka ko³owa, potrzebne jest znacznie bardziej zindywidualizowane podejœcie. W niniejszym opracowaniu przedstawiono próbê generalizacji wymogów dla takiego systemu diagnostycznego. Pierwszym za³o eniem jest to, e monitorowanie stanu powinno byæ adaptacyjne na poziomie przed przetworzeniem sygna³u, na poziomie ekstrakcji cech diagnostycznych i wreszcie na poziomie diagnostycznego wyci¹gania wniosków. Poni sze kwestie zosta³y przedstawione jako praktyczny wynik badañ prowadzonych w tej dziedzinie: adaptacyjny pre-procesor sygna³u (filtr adaptacyjny) dla wykrywania lokalnych uszkodzeñ skrzynek przek³adniowych i ³o ysk (G&B), adaptacyjny subsystem dla wykrywania uszkodzeñ lokalnych (G&B), adaptacyjny subsystem diagnostycznego wyci¹gania wniosków dla diagnostyki uszkodzeñ lokalnych (G), adaptacyjny subsystem diagnostycznego wyci¹gania wniosków dla oceny wspó³pracy kó³ zêbatych (G). Zostanie on zilustrowany przyk³adami opartymi na danych przemys³owych pobranych z maszyn górniczych podczas zwyk³ej (niestacjonarnej) pracy, a mianowicie z dwustopniowej skrzynki przek³adniowej stosowanej w napêdzie przenoœnika, z³o onej skrzynki przek³adniowej z uszkodzonym stopniem planetarnym (koparka ko³owa), ³o ysk stosowanych w ko³ach pasowych (przenoœniki taœmowe). ADAPTIVE APPROACHES FOR CONDITION MONITORING OF MINING MACHINES Key words Adaptive system, open pit mine, diagnostic technique Abstract Mining machines often work under non-stationary, Time-Varying Load Conditions (TVLC). The non-stationarity of load is mainly caused by the non-stationarity of technological process of mining (variation of external load caused by an operating bucket wheel, a time-varying stream of materials transported by conveyors, etc.). Taking into account variation of external load, complex mechanical structure of diagnosed object, enviromental impact, high level of internal/external interference, possibility of multi-faults appearance (in different location, development stage, nature of damage, i.e. distributed/localised, etc.) condition monitoring becomes a serious task. Indeed, diagnostics of wear, and especially of local damages (for example related to tooth crack/breakage) is complicated and unfortunately even proffesional diagnostics systems are sometimes not able to recognize it. In author s opinion one of possible reasons is that diagnostic systems are closed, with rigid rules of
14 116 signal processing and reasoning. For example for such unique machines like bucket wheel excavator there is a need for more individual approaches. In this paper an attempt of generalization of requirements for such diagnostic system is presented. First assumption is that condition monitoring should be adaptive at signal preprocessing level, diagnostic feature extraction level and finally, at diagnostic reasoning level. As a practical results of research carried out in this field following issues will be presented: adaptive signal pre-processor (adaptive filter) for local damage detection in gearboxes and bearing (G, B), adaptive subsystem for local damage detection (G, B), adaptive subsystem of diagnostic reasoning for local damage diagnostics (G), adaptive subsystem of diagnostic reasoning for geared wheel cooperation assesment (G). It will be illustrated by examples based on industrial data captured from mining machines during normal (non-stationary) operation, namely two stage gearbox used in driving unit for belt conveyor transportation, complex gearbox with damaged planetary stage (bucket wheel excavator), bearing used in pulleys (belt conveors).