Fuzzy PI Control of an Industrial Weigh Belt Feeder Yanan Zhao and Emmanuel G. Collins, Jr. Department of Mechanical Engineering FAMU-FSU College of Engineering August 2001 Department of Mechanical Engineering Florida State University Contents Introduction PI-like Fuzzy Logic Controller PI Fuzzy Logic Controller Conclusions Department of Mechanical Engineering Florida State University Introduction IIndustrial d t i l weigh i h belt b lt feeder from Merrick I d ti Industries, Inc. I Transports solid materials i t a manufacturing into f t i process at a constant rate. In current practice practice, the PI tuning is performed manually. An automated tuning process is desired. p Department of Mechanical Engineering Florida State University Introduction (cont (cont’d) d) The nonlinearities of the weigh belt feeder are: motor saturation (control signal with [0,10] volt motor friction friction, significant sensor quantization. M Model-based d l b d friction f i ti compensation ti methods th d have h limitations: characteristics of friction are difficult to analyze Fuzzy logic control (FLC) is a solution. Department of Mechanical Engineering Florida State University Introduction (cont (cont’d) d) Inference mechanism Rule-base Defuuzzification reference fuzzzification Fuzzy logic controller control signal Plant plant performance FLC is p particularlyy useful when the plant p model is unknown or difficult to develop. fuzzification the rule rule FLC has four main parts: fuzzification, base, the inference engine, and defuzzification. Fuzzy PID control has been widely studied studied. Department of Mechanical Engineering Florida State University Introduction (cont (cont’d) d) PID-like PID like ffuzzy logic cont controller olle (FLC) (FLC): Δu = F(e, Δe), u = F(e, Δe), Δu = F(e, Δe, Δ2e) The h structure is analogous l to that h off the h conventional PID controller. PID FLC FLC: u (k ) = (K p + K i z + K z −1 d )e(k ) The gains are tuned on-line with fuzzy reasoning. This requires more experience with the system. Both PI-like and PI FLCs are designed and implemented. Department of Mechanical Engineering Florida State University PI-like PI like Fuzzy Logic Control r e + - e(k) \ Δe(k) NB NM NS ZE PS PM NB NB NB NB NB NM NS NM NB NM NM NM NS ZE Δe NS NB NM NS NS ZE PS Ge GΔe NE NM NM NS ZE PS PM Δu PI like PI-like FLC u GΔu + + plant y 1/z PS NS NS ZE PS PS PM PM PB NS ZE ZE PS PS PM PM PB PM PB PM PB PB ZE PS PS PM PB PB PB Fuzzy Rules for Computation of Δu MFs of e, Δe, and Δu Department of Mechanical Engineering Florida State University PI-like PI like FLC(cont FLC(cont’d) d) Scaling g factors (SFs) ( ) appear pp as follows: eN = Gee, ΔeN = GΔeΔe, Δu = Gu ΔuN The SFs play a role similar to that of the gains of a conventional controller. Selection S l ti off the th SFs SF are b based d on expertt knowledge and adjustment rules d developed l d by b evaluating l ti the th control t l results. Department of Mechanical Engineering Florida State University PI-like PI like FLC(cont FLC(cont’d) d) For the weigh g belt feeder,, controllers were designed for setpoints of 1, 2, …, 5 volts. Constant scaling factors were used. The output SF needs to be tuned due to its strong influence on the performance and stability. Department of Mechanical Engineering Florida State University Gain Scheduling of PI PI-like like FLC(cont FLC(cont’d) d) Gain scheduling r e + - Δe Ge GΔe PI-like FLC Δu u GΔu + + plant y 1/z Adjust djust tthe e output sca scaling g factor acto us using g G Δu ,sp = G Δu , 0 ⋅ γ, γ = 1 1 + 0.1 ⋅ sp The control effort is decreased with the increasing of the setpoint. Department of Mechanical Engineering Florida State University Self-tuning Self tuning of PI PI-like like FLC(cont FLC(cont’d) d) α Fuzzy reasoning r e + - Δe Ge GΔe PI-like FLC Δu u GΔu + + plant y 1/z Adjust djust tthe e output sca scaling g factor acto as Δu = (α ⋅ G u ) ⋅ Δu N The updating factor α is tuned online based on fuzzy reasoning using the error and change of error at each sampling time. Department of Mechanical Engineering Florida State University Self-tuning of PI-like FLC(cont FLC(cont’d) d) MFss for o MFs for e(k) \ Δe(k) N ZE P Rule-bases for N B M S ZE M S M computation of P S M B The domain of the updating factor is also Department of Mechanical Engineering tuned. Florida State University Comparison of the gain scheduled and the self-tuning FLCs Setpoint Type of FLC IAE ISE ITAE ITSE 1 GS 523 8 523.8 372 2 372.2 3251 4 3251.4 811 6 811.6 ST 456.0 303.2 3114.4 585.3 GS 755.1 1032.0 4200.5 1625.4 ST 689.6 883.6 4161.0 1241.6 GS 1071.1 2192.9 5172.5 3132.6 ST 948.5 1952.1 4223.3 2440.7 2 3 C Comparison i Using U i Different Diff t Performance P f Indices I di The self-tuning PI-like FLC yields better performance. Department of Mechanical Engineering Florida State University Comparison of the gain scheduled and the self-tuning FLCs(cont’d) Comparison of the Performance at Setpoints of 1, 2 and 3 Volts Department of Mechanical Engineering Florida State University Comparison of the gain scheduled and the self-tuning FLCs(cont’d) Gain scheduled FLC changes only the range of the output p surface. Self tuning FLC changes Self-tuning both the range and the shape of the output surface. Department of Mechanical Engineering Florida State University PI Fuzzy Logic Control Fuzzy reasoning r e + - Δe Ge GΔe Fuzzy reasoning Ti Kp PI controller: H(z) = K p + K i PI Controller u plant y z 1 z ) = K p (1 + z −1 Ti z − 1 The proportional gain K p and integral time constant T i are adjusted on-line by fuzzy reasoning. Department of Mechanical Engineering Florida State University PI FLC (cont (cont’d) d) MFs of e(k) and Δe(k) MFs of Kp MFs of Ti Department of Mechanical Engineering Florida State University PI FLC (cont (cont’d) d) Fuzzyy rules for computation of Kp: e(k) \ Δe(k) N N B ZE S P B e(k) \ Δe(k) Fuzzy rules for computation of Ti: ZE B B B N ZE N S S ZE B M P S S P B S B P S B S For different setpoints the range of Kp is 1 ρ= adjusted. K = ρ ⋅ K 1 + 0.2 ⋅ sp p , max ma p , max ma 0 , MFs of Ti for different setpoints are also Department of Mechanical Engineering Florida State University adjusted. PI FLC (cont (cont’d) d) Experimental results at Setpoints of 1 1, 2 2, 3: Department of Mechanical Engineering Florida State University Comparison of the self self-tuning tuning PI PIlike FLC and PI FLC S e tp o in t T y p e o f FLC 1 2 3 IA E IS E IT A E IT S E PI 3 1 7 .6 1 4 9 .8 2 6 1 9 .8 2 3 9 .4 ST 4 5 6 .0 3 0 3 .2 3 1 1 4 .4 5 8 5 .3 PI 5 0 1 .6 5 9 0 .6 2 9 4 3 .5 5 9 1 .0 ST 6 8 9 .6 8 8 3 .6 4 1 6 1 .0 1 2 4 1 .6 PI 7 5 9 .6 1 5 2 2 .1 3 7 9 8 .5 1 5 9 2 .1 ST 9 4 8 .5 1 9 5 2 .1 4 2 2 3 .3 2 4 4 0 .7 The PI FLC performed better. Department of Mechanical Engineering Florida State University Co c us o s Conclusions T Two categories t i off ffuzzy PI controllers t ll were designed: gain-scheduled PI-like FLC, self-tuning PI-like FLC PI FLC with g gains tuned byy fuzzyy reasoning g The self-tuning PI-like FLC performed better than the gain scheduled PI-like FLC FLC. The PI FLC performed better than the two PI-like FLCs. Department of Mechanical Engineering Florida State University Conclusions(cont d) Conclusions(cont’d) As more user knowledge is incorporated into the controller design, the performance of the FLC improved. i d proposed p are quite q simple, p All of the rules p making the methods suitable for implementation in an industrial environment. Department of Mechanical Engineering Florida State University