Inter national J our nal of Electrical and Computer Engineering (IJECE) V ol. 11, No. 2, April 2021, pp. 1697 1708 ISSN: 2088-8708, DOI: 10.11591/ijece.v11i2.pp1697-1708 r 1697 Monitoring of solenoid parameters based on neural netw orks and optical fiber squeezer f or solenoid v alv es diagnosis Abdallah Zahidi 1 , Said Amrane 2 , Nawfel Azami 3 , Naoual Nasser 4 1,2,3 INPT Optics Lab, National Institute of Posts and T elecommunications, Rabat, Morocco 4 LDEDS, F aculties of Science and T echnology , Hassan Uni v ersity 1 st , Settat, Morocco Article Inf o Article history: Recei v ed Apr 4, 2020 Re vised Aug 11, 2020 Accepted Sep 30, 2020 K eyw ords: EMS Fluctuations Monitoring Neural Netw orks Polarization squeezer ABSTRA CT As crucial parts of v arious engineering systems, solenoid v alv es (SVs) operated by electromagnetic solenoid (EMS) are of great importance and their f ailure may lead to cause une xpected casualties. This f ailure, characterized by a de gradation of the per - formances of the SVs, could be due to a fluctuations in the EMS parameters. These fluctuations are essentially attrib uted to the changes in the spring constant, coef ficient of friction, induc tance, and the resistance of the coil. Pre v enti v e maintenance by con- trolling and monitoring these parameters is necessary to a v oid e v entual f ailure of these actuators. The authors propose a ne w methodology for the functional diagnosis of electromagnetic solenoids (EMS) used in h ydraulic systems. The proposed method monitors online the electrical and mechanical parameters v arying o v er time by using articial neural netw orks algorithm coupled with an optical fiber polarization squeezer based on EMS for polarization scrambling. First, the MA TLAB/Simulink model is proposed to analyze the ef fect of the parameters on the dynamic EMS model. The result of this simulation i s used for training the neural netw ork. Then a simulation is proposed using the neural net tting toolbox to determine the solenoid parameters (Re- sistance of the coil R, stif fness K and coef ficient of fric tion B of the spring) from the coef ficients of the transfer function, established from the model step response. Future w ork will include not only diagnosing f ailure modes, b ut also predicting the remaining life based on the results of monitoring. This is an open access article under the CC BY -SA license . Corresponding A uthor: Abedallah Zahidi National Institute of Posts and T elecommunications A v enue Allal Al F assi, Rabat, Morocco Email: zahidiabdo@yahoo.fr 1. INTR ODUCTION In electromechanical solenoids (EMS), also called electromagnetic de vices (EMD), dri v en by a control solenoid, the armature is positioned by balancing the electromagnetic force ag ainst that of a return spring. The y are relati v ely an ine xpensi v e construction [1], ha v e a simple design and control circuit, require li ttle ener gy for control, are highly reliable [2], these electromagnetic actuators are used in wide range of modern industrial equipment such as digital actuator arrays [3], v ehicle vibration control systems [4], g as v alv e [5], robotic ma- nipulators [6], positioners [7], anti-braking systems [8], and polarization controllers where solenoids are used as mechanical actuator on the fiber to adjust the output light po wer [9]. In h ydraulic systems, these electro- magnetic actuat ors are comm o nl y used in three basic cate gories to actuate h ydraulic control v alv es directional- J ournal homepage: http://ijece .iaescor e .com Evaluation Warning : The document was created with Spire.PDF for Python.
1698 r ISSN: 2088-8708 control, flo w-control, and pressure-control. Directional-control v alv es are used to connect and isolate h ydraulic passages by simply opening a nd closing a communication path. Flo w-control v alv es allo w v ariable flo w rate control to a component. Finally , pressure-control v alv es re gulate v ariable pressure to a h ydraulic component [10]. Analysis of EMS purposes sho w that to impro v e the safety and the performance of the EMS based h ydraulic system, t he re gular maintenance strate gy of these electromechanical solenoids surely reduces f ailure rate of the system ; ho we v er , it brings high maintenance cost [11]. It should be highlighted that emer genc y modes of EMDs a re not only results of f aults of their v arious elements or incorrect personnel acti vities during EMD manuf acturing and operation, the y are also possible during a normal operation due to the wearing of mating surf aces of friction assemblies or/and fluctuation in its parameters [12, 13]. Therefore, it is necessary to de v elop an ef fecti v e approach for EMS diagnostics and operation controls in order to map the f ailure and the reliability of the EMS based on the solenoid v alv es. The diagnostic results could be used to estimate the health condition of the solenoid v alv es and predict their remaining useful life. Additionally , the s olenoid v alv es coul d get tim ely repaired or replaced before their potent ial f ailure causes an y system breakdo wn. 2. PR OBLEM ST A TEMENT Despite the high reliability and the e xcellent performance of solenoid v alv es in v arious appli cations, their f ailure may result in se v ere system crash, signficant casualties and economic losses, especially in safety- first fields, such as rail w ay braking system, a viation engine, and nuclear po wer plants [14]. Ho we v er , as stated in some pre vious w orks [15, 16], the EMS dynamic in these de vices is go v erned by an electromagnetic force that increases greatly when the air g ap is near zero. This nonlinear beha vior , together with ph ysical bounds that limit the motion, causes EMS of v alv es to be s ubject to strong shocks and wear that often result in early f ailures [17], these f ailures, which af fects the dynamic response [18], may be due to a fluctuation in the EMS structure parameters [19] or the electrical and mechanical parameters of the EMS (v ariations in inductance and resistor of the coil, changes in spring constant and coef ficient of friction) [20, 21]. Other studies ha v e sho wn a deterioration in the performance of EMS based system in presence of parametric v ariation [12, 13], e v en for the best control solutions [20]. Other literatures ha v e also re v ealed that f ailure of solenoid v alv es can also occur gradually due to coil b urnout of EMS related to mains v oltage and frequenc y , spring force [22, 23]; and that the resistance of the coil might be a source of themo-mechanical f ailure of the solenoid v alv e [24]. In addition, the abo v e-mentioned literature pro vides the parameters which characterize the f ailure of the solenoid v alv es. This information can then be used to design a m o de l to map the f ailure. Approaches based on signal processing [25] and machine learning [26] ha v e been proposed in the lit erature to diagnose the state of solenoid v alv es leading to the de v elopment of a sensor to detect anomalies [27] or a method for grouping f ailures [26]. None of these approaches gi v es a ph ysical e xplanation of the f ailure modes related to the solenoid parameters [28]; Other models based on the mo v ement of the armature and F oucault current [2] and the EMD winding current curv e appearing with the mo v ement of the armature [29, 30] ha v e been de v eloped for diagnosis, b ut none of these models does treat electrical and mechanical parameters as a source of f ailure. More v er , most control approaches are using signals such as the coil current or v oltage of solenoid to monitor the parameters; yet, the main problems in such approaches are that the detected signals ar e prone to interference and difcult to obtain [31]. Other w orks ha v e been limited to the estimation and identication of a single solenoid parameter [32, 22]. In this w ork, the authors ha v e de v eloped a ne w approach for the diagnosis of an EMS actuator in solenoid v alv e. The proposed approach is based on a ne w method based on optical fiber polarization controller signal feedback coupled with articial neural netw orks model (ANN) for monitoring the electrical and mechanical EMS parameters (resistence of coil R, and K, B respecti v ely the stif fness and the coef ficient of friction of the spring), considered in this approach as health indices to characterize the f ailure of the solenoid v alv e. NN has the adv antages of controlling comple x and non-linear systems [33, 34], has high accurac y of prediction capability [35], and it is of great importance to find the high speed electromagnetic switching v alv e [36]. 3. METHODOLOGY OF STUD Y This paper proposes a ne w methodology using optical fiber polarization controller signal feedback coupled with an articial neural netw orks model (ANN) for monitoring the solenoid parameters and predicting its performance. Figure 1 sho ws the proposed monitoring process. Int J Elec & Comp Eng, V ol. 11, No. 2, April 2021 : 1697 1708 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Elec & Comp Eng ISSN: 2088-8708 r 1699 Laser 1550nm EMS of solenoid v alv e Sensor of armature pressure (Fiber) Analyzer Detector (photodiode) Neural net- w ork model Storage & vie wing of EMS parameters Figure 1. Block diagram of monitoring process In the proposed struct ure, the optical fiber polarization controller is based on an EMS and an opti cal fiber used as a mechanical force sensor of the EMS armature. This force induces an optical birefringence that modifies the polarization of the light [9]. The v ariation of the polarization of the light is reflected by the v ariation of the light intensity detected by a photodiode placed at the output of the optical fiber , In the first step, a mathematical model is proposed to obtain the response of the system. Then, this model is identied from this response using function tfest of MA TLAB/Simulink [37] to determine the transfer function coef ficient of the system. In the second step, the ef fect of solenoid parameters v ariation on the transfer function coef ficient is analysed. This method uses simulation electromagnetic fiber squeezer based polarization controller with function tfest and the simulation results are stored in a te xt file that will be used for neural netw orks training. In the last step, the neural netw orks model is proposed to estabish the solenoid parameters from the coef ficients of the transfer function set from the step response of the fiber squeezer . Finally , to check the ef ficienc y of the proposed model, a prediction error is calculated. The result of the simulation sho ws that this optical fiber squeezer coupled with the neural netw ork model is v ery ef ficient to monitor the EMS parameters. The results of monitoring will be used in a future w ork to estimate the remaining life of the solenoid v alv e. 4. B UILDING THE SIMULINK MODEL 4.1. Structur e and equations of electr omagnetic fiber squeezer The EMS is the electromagnetic actuator that e x erts the pressure on the fiber . Its structure is sho wn in Figure 2. Analyser Detector Laser 1550 Solenoid Figure 2. Scheme of using the electromagnetic fiber squeezer The magnitude of the phase dif ference of tw o polarized light along the squeezing axis and its orthog- onal axis can be e xpressed as [38]: = 6 e 5 F d (1) and the light po wer P s at the output of the polarization analyzer according to scheme of Figure 2 is: P s = E 2 (2) Monitoring of solenoid par ameter s based on neur al networks and optical fiber ... (Abedallah Zahidi) Evaluation Warning : The document was created with Spire.PDF for Python.
1700 r ISSN: 2088-8708 where E = A: 1 2 :E (3) A =   e j ( m + 2 ) 0 0 e j ( m 2 ) ! (4) P s = E 2 = P 0 2 (1 + cos ) (5) where P 0 is the input light intensity . 4.2. Mathematical model of EMS The solenoid refers to the electromagnetic actuator . It is used to e x ert pressure on the fiber . Its struc- ture is sho wn in Figure 3. Plunger Coil Figure 3. Cross section of EMS The mathematical model of EMS is gi v en [39] by e xpression as (6): m d 2 x ( t ) dt 2 + B dx ( t ) dt + K x ( t ) = 0 r N 2 AI 2 ( t ) 2( x 0 x ( t )) 2 (6) where: x ( t ) : Displacement of the armature in (m), I ( t ) : The electromagnet coil current in(A), A : the cross sectional area of the coil in (m 2 ), N : the number of the turns of the coil, 0 : P ermeability of the free space in (H/m), r : Relati v e Permeability of the die lectric materiel between the coil and armature, x 0 : The initial air g ap between the armature and the backside of the frame in (m), m : Masse of the armature in (Kg), K : is the stif fness of the spring in (N/m) and B : System damping coef ficient in (N.s/m). The equation of the electrical circuit is as (7) and (8): u = R i ( t ) + d dt [ L ( x ) :i ( t )] (7) u = R i ( t ) + L ( x ) di ( t ) dt + i ( t ) dL ( x ) dt (8) R is the series resi stance of the EMS coil and L ( x ) is the inductance of the coil that depends of the air g ap [39]: L ( x ) = 0 r x 0 x ( t ) (9) The balance equation of the force acting on the fiber is e xpressed as [40] m d 2 x ( t ) dt 2 = F K ( x ( t ) x 0 ) B dx ( t ) dt (10) F is the force produced by the magnetic field and it be deri v ed kno wing that magnetic system is linear and that current w as k ept constant F = dw t dx = i 2 2 dL ( x ) dx = i 2 2 aL 0 ( a + x ) 2 (11) Int J Elec & Comp Eng, V ol. 11, No. 2, April 2021 : 1697 1708 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Elec & Comp Eng ISSN: 2088-8708 r 1701 where L 0 = 0 adN 2 g and a , d , g parameters depending on the EMS. from (8) we can write di dt = 1 L ( x ) u R i i dL ( x ) dx dx dt (12) from (10) we can write d 2 x dt 2 = 1 m F K ( x x 0 ) B dx dt (13) Both (1) and (5) are used to write: P s = P 0 2 1 + cos ( 6 e 5 F d ) (14) 4.3. Simulink model The system model has been implemented in v ersatile softw are MA TLAB which is widely used in control engineering around the w ord. This simulation is used to ef fecti v ely determine the best performance of the dynamic response in the output light intensity . The electrical model Simulink which models (12) is represented in Figure 4(a); while, the Simulink mechanical model for (13) is illustrated in Figure 4(b). This model depends on the intrinsic parameters of the EMS: the mass m of the armature, the coef ficient of friction B, the stif fness of the ress ort K, the resistance and the inductance of t he coil (R, L). The Figure 4(c) represents the optical Simulink model for (14). (a) (b) (c) Figure 4. Simulink models, (a) Electrical model, (b) Mechanical model, (c) Optical model Monitoring of solenoid par ameter s based on neur al networks and optical fiber ... (Abedallah Zahidi) Evaluation Warning : The document was created with Spire.PDF for Python.
1702 r ISSN: 2088-8708 5. PRINCIPE OF SIMULA TION 5.1. Monitoring pr ocess flo wchart T o predict and monitor the EMS parameters, the coef ficients of the transfer function obtained from the step response of the fiber squeezer based on the EMS are used as input of the NN model. The parameters solenoid R, B and K are e xpected as outputs . The flo wchart of the monitoring process is sho wn in Figure 5. Figure 6 proposes the architecture of the ANN model and T able 1 illustrates an e xample of identification results used for the ANN training. Figure 5. Monitoring process flo wchart Input 3 Hidden lay er 10 w b + Output lay er 3 w b + Output 3 Figure 6. Neural netw ork architecture T able 1. Identification result for neural netw ork training Indice K (N/m) B(Ns/m) R( ) Num Dun1 Dun2 Dun3 1 2998.5692 3.2177 11.8536 114.7376 1 16.2331 15095.4567 2 2168.9635 2.3559 12.8581 70.5311 1 11.8918 10931.6472 3 2521.0054 2.1333 13.6778 72.4483 1 10.7603 12681.9928 5.2. Neural netw orks ar chitectur e The NN inputs consist of a matrix of order 10000 3 , where each line represents a set of coef ficients of the transfer function num, dun2 and dun3. On the other hand, the outputs are the elements of a matrix of the same order as the inputs where each line e x emplifies a set of solenoid parameters (K, B and R). The structure also contains 10 hidden layers chosen by def ault with the sigmoid function as act i v ation function and 3 output layers with a linear acti v ation function as sho wn in Figure 6. 5.3. Neural netw orks training 5.3.1. Identification with v ariable parameters f or NN training First, we propose a mathematical model that is used to obtain the response of the system. Then, this model is identified from this response using identification function tfest whose syntax is sys=tfest(data, np, nz). This function is used to estimate a transfer function containing nz zeros and np poles from the inde x response Int J Elec & Comp Eng, V ol. 11, No. 2, April 2021 : 1697 1708 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Elec & Comp Eng ISSN: 2088-8708 r 1703 (data) of the Simulink model described in paragraph 4.3. The resulting transfer function has the coef ficient (num, dun2, dun3), and can be e xpressed as (15): T f = num s 2 + dun 2 s + dun 3 (15) Secondly , in order to obtain the system transfer function for dif ferent v alues of the solenoid parame- ters, K, B and R are v aried while while k eeping m=200 g and L=20 mH since the y are not lik ely to v ary during long-term operation of the solenoid. The v ariation interv al of K is 1000 to 3000 N/m ; R is 10 to 15 and B is 2 to 4 N.s/m. The inputs of the Simulink model are a matrix of three columns and n ro ws of random v alue of solenoid parameter . The random v ariation between the max and min v alues of each of the three parameters (K, B and R) is obtained by using the function rand (n,1) whose syntax is as (16): par ameter = ( max { min ) r and ( n; 1) + min (16) where parameter is K, R or B, and n= 10 000. The follo wing Figure 7 represents the flo wchart which allo ws to obtain the dataset and T able 1 sho ws an e xample of the te xt file results obtained. The results of this simulation are used as data (num, dun2, dun3) and tar get (K, B, R) for training the neural netw ork model as sho wn in section 5.3.2. Start initialization K, B and R te xt files Opening Sumlink model Identification with tfest function Coef Tf* te xt files V ariation of EMS parameters (K, B, R) End VSP* End Tf: T ransfer function VSP: V ariation solenoid parameters Figure 7. Identification flo wchart 5.3.2. Neural netw orks training The tar get of neural netw ork is able to identify and predict the solenoid parameters. Data from step response and neural netw ork tar get are used to search weight (w) and bias (b). W eight and bias are obtained by entering data and tar get in MA TLAB program by using Neural Net fitting toolbox which of fers se v eral training functions. The updating of weight and bias v alues during netw ork training is performed according to the Le v enber g-Marquardt optimization which of fers f aster tracking of system parameter change [41]. The Le v enber g-Marquardt algorithm is an ef ficient and popular damped least squre technique. This algorithm is a combinaison between the steepest gradient descent and the Gauss-Ne wton algorithms [42]. The acti v ation function at the output of the HLs is the sigmoid function, it deli v ers a continuously smoother range of v alues between 0 and 1 and is less e xpensi v e in terms of calculation. At the output of the netw ork, the acti v ation function is linear , which creates an output signal proportional to the input. During the searching process of weight and bias, the dataset is subdi vided into three percent, 70% for training, 15% for the test and 15% for Monitoring of solenoid par ameter s based on neur al networks and optical fiber ... (Abedallah Zahidi) Evaluation Warning : The document was created with Spire.PDF for Python.
1704 r ISSN: 2088-8708 v alidation. The e v aluation of the model is measured using three e v aluation performances which are the mean square error (MSE), the coef ficient of correlation (R) and error histogram. The optimization technique applied in the ANN model training seeks to optimize the we ights and biases of the ANN structure by minimizing rhe mean square error (MSE). The training results are ill ustrated in Figure 8(a) which represents the con v er gent curv e of the MSE according to the epochs, error histogram and coef ficient of the correlation between the output and the tar get that respecti v ely illustrated in Figure 8(b) and Figure 8(c). 0 200 400 600 800 1000 1000 Epochs 10 -4 10 -2 10 0 10 2 10 4 10 6 Mean Squared Error  (mse) Best Validation Performance is 6.6317e-05 at epoch 1000 Train Validation Test Best (a) 0 2000 4000 6000 8000 10000 12000 Instances Error Histogram with 20 Bins -0.06544 -0.05807 -0.0507 -0.04333 -0.03596 -0.02858 -0.02121 -0.01384 -0.00647 0.000902 0.008274 0.01564 0.02302 0.03039 0.03776 0.04513 0.0525 0.05987 0.06724 0.07462 Errors  Zero Error (b) 500 1000 1500 2000 2500 Target 500 1000 1500 2000 2500 Output ~= 1*Target + -1.9e-06 Regression R=0.99 Y=T Fit Data (c) Figure 8. T raining performance, (a) The con v er gent curv e of the MSE, (b) T raining error histogram, (c) T raining coef ficient of correlation 5.4. Monitoring r esult and pr ediction err or The prediction and monitoring, of the EMS parameters are achie v ed through the data acquisition (num, dun2, dun3) from the step response of the optical fiber polarization controller based on EMS. These data are used as inputs of the NN model in order to find the parmeters. The predicted solenoid parameters M, B and K obtained from the transfert function coef ficient are illustred in the T able 2. Figure 9 sho ws the structure of the minotoring process. Int J Elec & Comp Eng, V ol. 11, No. 2, April 2021 : 1697 1708 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Elec & Comp Eng ISSN: 2088-8708 r 1705 T able 2. Predicted parameters testing result NN input Solenoid parameter num dun2 dun3 K(N/m) B(Ns/m) R( ) T est1 74.23 18.28 5683 1101.8 3.60 8.90 T est2 137.86 14.03 14010 2775.5 2.76 10.40 T est3 59.12 16.21 10871 2159.5 3.22 14.03 Figure 9. The model architecture for EMS monitoring process 5.5. Model perf ormance testing The e v aluation of the model performance is done according to the flo wchart of the Figure 10 by using data not utilized for the model training. The parameters found are used as inputs of the Simulink model to find the predicted v alue of transfer function coef ficients. These v alues are compared to the current v alues :num, dun2 and dun3 to finally calculate the prediction error . The model performance t esting structure is illustrated in Figure 11. Figure 10. Performance testing flo wchart Monitoring of solenoid par ameter s based on neur al networks and optical fiber ... (Abedallah Zahidi) Evaluation Warning : The document was created with Spire.PDF for Python.
1706 r ISSN: 2088-8708 Figure 11. The detail model performance testing 5.6. Results and discussion T able 1 illustrates the result of the mathematical model identification of the system proposed for training using the tfest function for dif ferent v alues of the parameters K, B and R. It is clear from this table that if the stif fness of t h e spring K, the coef ficient of friction B and the resistance of the solenoid coil R v ary , the coef ficients of the transfer function characterizing the model dun2 and dun3 also v ary , this v ariation af fects the performance of the h ydraulic system based on solenoid v alv e. Ho we v er , the coef ficient dun1 is k ept constant it is equal to 1. The Figure 8(a) sho ws the con v er gence curv e of MSE which illustrates the e v olution of the NN training. Moreo v er , it can be observ ed from this figure that suitable weights and biases of the NN model are found in the end of the iteration with a better MSE v alue which is around 6,63.10 6 . In addition, the impro v oment of the model performance might be realized by increasing the number of iterations (epochs) in order to minimize the MSE. The Gaussian form of the error histogram in Figure 8(b) and the v alue of coef ficient of correlation between outputs and tar gets during the test illustrated in Figure 8(c) sho w the high quality of training result. On the other hand, according to the results obtained from the performance test model of Figure 9 along with those illustrated on the T able 2, it is clear that the predicted v alues of K, B and R are in e vident agreement with the test v alues obtained by the model of the Figure 11 and which are illustrated in T able 3. What’ s more, and generally , the training result is less accurate compared to the training, ho we v er , from the testing performance gi v en in the T able 3 and the relat i v e prediction error illustrated in T able 4. Thus it is pro v en that the training performance of Le v enber g-Marquardt algorithm is able to control and predict the solenoid parameters, and that the proposed neural netw ork monitoring w as successfully implemented. T able 3. Simulink model testing result Solenoid parameters Simulink model outputs K(N/m) B(Ns/m) R( ) num dun2 dun3 T est1 1101.8 3.60 8.90 74.73 18.34 5684 T est2 2775.5 2.76 10.40 137.74 14.05 14010 T est3 2159.5 3.22 14.03 58.97 16.21 10870 T able 4. The test result prediction error Actual v alues (NN input) Predicted v alues (simulink model output) Relatif prediction error num dun2 Dun3 num dun2 dun3 num dun2 Dun3 T est1 74.23 18.28 5883 74.73 18.34 5684 0.67% 0.30% 0.01% T est2 137.86 14.03 14010 137.74 14.05 14010 0.08% 0.13% 0% T est3 59.12 16.21 10871 58.97 16.21 10870 0% 0% 0.003% 6. CONCLUSION In this article, a model-based approach for detecting f aults in electromagnetic solenoi d of v alv es w as proposed. The model is based on ANN coupled with an optical fiber polarization squeezer signal feedback. During a v oltage step, the coef ficients of the transfer function of the mathematical model are determined from the step response of the actuator model. Then, the parameters of the EMS, sele cted as the health indices of the solenoid v alv e, are determined from the ANN model. The results of this model ha v e been v erified through simulation on MA TLAB/Simulink. The propose d neural netw orks model has satisf actory performance of prediction and has met the monitoring requirement. Thus, this contrib ution pro vides a no v el approach on f ault diagnostics in h ydraulic systems. It consists only of softw are and optical fiber . Additionally , e xpensi v e Int J Elec & Comp Eng, V ol. 11, No. 2, April 2021 : 1697 1708 Evaluation Warning : The document was created with Spire.PDF for Python.