In a fuzzy inference processor having a fuzzification unit which, for each input variable, emits a minimum linguistic value number and a maximum linguistic value number and values of the membership functions of linguistic values that are in a closed interval between said minimum and maximum linguistic value numbers, a knowledge base memory which emits numbers for linguistic values of each input variable prescribed in a plurality of rules stored in said knowledge base memory, and a defuzzification unit that conducts defuzzification dependent on weighting factors applied to a linguistic value of an output variable, the improvement comprising:
- a rule decoder supplied with said minimum and maximum linguistic value numbers, a selection signal former supplied with said minimum and maximum linguistic value numbers, and a rule evaluation unit which emits said weighting factors;
- said knowledge base memory comprising means for emitting, for each input variable, a linguistic value for said input variable prescribed by said rules;
- said rule decoder comprising means for generating a hit signal indicating whether, and if so which of, said rules are fulfilled by said input variable and, if necessary, a mask signal indicating that said input variable does not occur in any of said rules, said hit signal being supplied to said selection signal former and said mask signal being supplied to said rule evaluation unit;
- said selection signal former comprising means for forming a selection signal dependent on said hit signal, said minimum and maximum linguistic value numbers, and said rule word and for emitting said selection signal to said rule evaluation unit;
- said rule evaluation unit comprising a plurality of first hold elements, clocked by a first clock signal and respectively supplied with said values of said membership functions from said fuzzification unit, an allocation unit, a plurality of two-input OR gates, a plurality of minimum circuits, a plurality of second hold elements clocked by a second clock signal, a plurality of drivers, a counter, a read/write memory addressable by said counter, and a multiplexer, the pluralities of first hold elements, OR gates, minimum circuits, second hold elements and drivers being equal, said plurality of first hold elements comprising, in combination, means for buffering said values of membership functions, one value per first holding element, and for transferring said values of said membership functions to said allocation unit upon said first clock signal, said allocation unit comprising means for allocating said values of said membership functions to one input of a selected one of said OR gates dependent on said selection signal, the other input of each OR gate being supplied with said mask signal, each minimum circuit having a first input connected to an output of one of said OR gates and a second input connected to an output of one of said second holding elements and comprising means for forming a running minimum from said first and second inputs and for supplying said running minimum to one of said drivers, each driver having an output connected to said read/write memory for storage of said running minima therein at respective addresses set by said counter, and also connected to said multiplexer and to one of said second holding elements, said plurality of second holding elements comprising, in combination, means for reading out running minima for a preceding clock cycle of said second clock signal from said read/write memory, one running minima per second holding element, and for transferring said minima from said preceding clock cycle respectively back to said second inputs of the respective minimum circuits, and said multiplexer comprising means for controlled individual emission of said outputs of said drivers as current weighting factors;
- means for emitting, to said rule evaluation unit, said linguistic values of said output variables;
- and
- said rule evaluation unit further comprising means for aggregating said current weighting factors, for supply to said defuzzification unit, by linguistically ORing current weighting factors for rules having a same linguistic value as said output variable.