Fuzzy logic

• 1. Introduction

  ›

  Fuzzy logic begins with a simple experiment: ask a group of people to name a color, and the answers spread across a range instead of landing on one word. Classical logic has no room for that spread. It only allows a statement to be true or false, nothing in between. Fuzzy logic instead lets a statement be partly true, somewhere between completely true and completely false. How do engineers build a working system around an idea that slippery? Where did that idea come from, and how did it end up quietly running real machines? This documentary follows fuzzy logic from its mathematical roots into the equipment it now controls.

• 2. Degrees, Not Absolutes

  ›

  A temperature reading feeding an anti-lock braking system might pass through several separate membership functions. Each one maps that same value onto a scale from 0 to 1. Boolean logic only allows the integers 0 or 1 for any variable, a strict on or off switch. Fuzzy logic instead allows any real number in between. That number captures how much a value belongs to a description, rather than simply whether it does.

  Degrees of truth and probabilities both sit on that same 0 to 1 scale, which makes them easy to confuse. They model different things. A degree of truth measures vagueness: how well a description like warm fits a given temperature. A probability measures ignorance: how likely an event is to happen at all. That distinction is what let fuzzy logic branch out far beyond pure mathematics, into control systems and, eventually, artificial intelligence.

• 3. A History Before the Name

  ›

  The term fuzzy logic entered use in 1965, when mathematician Lotfi Zadeh proposed fuzzy set theory. The underlying idea reached back further than that name suggests. Basic fuzzy logic had already been studied since the 1920s, framed as infinite-valued logic. Łukasiewicz and Tarski were notable figures in that earlier work.

  Zadeh's own contribution deepened through the 1960s and 1970s, alongside Joseph Goguen. Together they took up questions like linguistic variables and lattices. Fuzzy logic rests on a simpler observation than any of that math: people make decisions from imprecise, non-numerical information every day. Fuzzy models and fuzzy sets exist to represent that vagueness mathematically, recognizing and interpreting data that lacks certainty.

  That same drive to formalize imprecision would soon need a way to talk about words, not just numbers.

• 4. Turning Words Into Numbers

  ›

  A linguistic variable like age might take values such as young or its opposite, old. Natural language rarely contains enough distinct words to describe a full scale of truth. Fuzzy logic borrows adjectives and adverbs to stretch that vocabulary. Hedges such as rather and somewhat combine with a base term to produce values like rather old or somewhat young.

  One illustration maps a single temperature onto three functions at once, labeled cold, warm, and hot. A red arrow pointing to zero shows the temperature has no membership in hot. An orange arrow pointing to 0.2 gives it 0.2 membership in warm. A blue arrow pointing to 0.8 gives it 0.8 membership in cold. That process, assigning a numerical input to fuzzy sets with some degree of membership, is called fuzzification. A degree of 0 means the value does not belong in a set at all, and a degree of 1 means it belongs completely.

  Some fuzzy sets are defined with a sigmoid curve, including the standard logistic function, which carries its own built-in symmetry. Others are simply drawn as triangle or trapezoid shaped curves, each with a rising slope, a peak held at 1, and a falling slope.

  Those curves become the raw material a rule-based system consumes to reach a single decision.

• 5. Rules That Decide

  ›

  Engineers gave the name Mamdani to the most widely used fuzzy system, built on three steps. First it fuzzifies every input value into a fuzzy membership function. Then it executes every applicable rule in the rulebase to compute fuzzy output functions. Finally it defuzzifies those output functions into one crisp value.

  Fuzzy logic swaps in replacements for the Boolean gates AND, OR, and NOT. One common substitution defines AND(x, y) as the minimum of x and y, OR(x, y) as the maximum, and NOT(x) as 1 minus x. For true and false, coded as 1 and 0, these fuzzy operators produce exactly the same results as the Boolean versions. A second family of operators is built on multiplication instead. Here, x AND y equals x times y, and x OR y equals x plus y minus their product. Given any two of AND, OR, and NOT, the third can be derived from them. The generalized version of AND is known as a t-norm. A paper by Zaitsev and colleagues later set out a criterion for this: it recognizes whether an arbitrary choice table even qualifies as a fuzzy logic function at all.

  An IF-THEN rule might read: if temperature is very cold, then fan speed is stopped. If it is warm, fan speed is moderate; if hot, fan speed is high. When several rules point their THEN clause at the same output variable, their IF-side truth values are combined with the OR operator. Turning that mix of truth values back into one usable number is called defuzzification, and no single algorithm handles it. A common method cuts each output's membership function at its computed truth value. It then combines the resulting curves with the OR operator. Finally, it locates the center of weight of the area under that combined curve. The horizontal position of that center becomes the final output.

  A different school of fuzzy engineering builds that defuzzification step directly into the rules themselves.

• 6. The Takagi-Sugeno Approach

  ›

  Tomohiro Takagi and Michio Sugeno introduced the system that carries their names, for fuzzy identification of systems, modeling, and control. Sugeno and Kang later built on it with methods for identifying the structure of such fuzzy models directly from input-output data.

  The Takagi-Sugeno-Kang system, TSK for short, resembles Mamdani in its front end. It folds defuzzification into the execution of the rules themselves. Its consequent is written as a polynomial function instead of a fuzzy set, constant in a zero-order model or linear in a first-order one. A constant example might read: if temperature is very cold, then fan speed equals 2. There, the output simply matches that constant. A linear example might read: if temperature is very cold and humidity is high, then fan speed equals 2 times temperature plus 1 times humidity. Here, the antecedent's membership value acts as the rule's weight. Across an entire rule base, the final output becomes a weighted average of every rule's consequent.

  TSK's main advantage is computational efficiency, especially inside methods like PID control. It also works well with optimization algorithms and adaptive neuro fuzzy inference systems, and it guarantees a continuous output surface. Mamdani remains more intuitive and easier for people to work with directly. That is why TSK tends to stay embedded inside more complex systems rather than standing alone. A fuzzy system's output reflects a consensus of every input and every rule at once. That is why it keeps working reasonably well even when some inputs are missing or untrustworthy. Individual rules can carry weightings tied to their priority, reliability, or consistency. Those weightings can stay fixed, or shift dynamically based on other rules' output.

  That resilience under missing or unreliable data is what carried fuzzy logic out of research papers and onto real machines.

• 7. Trains, Vacuums, and Everyday Machines

  ›

  One widely cited fuzzy logic rule reads: "if you are close to the destination station and moving fast, increase the train's brake pressure." That kind of vague human instinct becomes something a control system can act on directly. Rules like that let engineers numerically refine vague human judgment, without forcing it into a rigid formula.

  The first notable fuzzy logic application ran on the Sendai Subway 1000 series, where fuzzy control improved the train's economy, comfort, and precision. Many of the earliest successful fuzzy logic applications, like this one, were built in Japan. The same approach went on to handle handwriting recognition in Sony's pocket computers and helicopter flight aids. It also reached other subway system controls, automobile fuel efficiency, single-button washing machine controls, and automatic power regulation in vacuum cleaners. Japan's Institute of Seismology, part of its Bureau of Meteorology, even used fuzzy logic for early earthquake recognition.

  That same appetite for handling uncertain, real-world signals is what eventually pulled fuzzy logic into conversation with an entirely different technology, the neural network.

• 8. Kin and Rivals

  ›

  Researchers in the 1980s split over which machine learning approach would win. One side backed decision tree learning, built on binary logic that matched the hardware running it. The other backed neural networks. Decision trees never delivered genuinely intelligent systems, despite serious effort. Neural networks, by contrast, produced accurate models of complex situations and spread across a wide range of electronic devices. When analyzed closely, the two fields turn out to share a deeper connection: the underlying logic of a neural network is itself fuzzy, not binary. A neural network takes a range of valued inputs, weights them against each other, and combines intermediate values repeatedly to reach a decision. None of that follows the either-or steps of non-fuzzy computer programming. Neural networks can now run directly on analog microchips, instead of the pseudo-analog approach older digital chips relied on, and that efficiency gain outweighs analog's lower inherent accuracy.

  One line captures the challenge starkly: "the envelope of what can be achieved and what cannot be achieved in medical diagnosis, ironically, is itself a fuzzy one." Medical and healthcare data is often subjective, which makes fuzzy logic an important tool in medical decision making. It has been applied to medical image analysis, biomedical signal analysis, and the segmentation and feature extraction of images or signals. The hardest question in this space is how much useful information fuzzy logic can actually extract, especially once the data has to come directly from patients. Image-based computer-aided diagnosis, tools built to help physicians reach a decision, has become one of the more common places fuzzy logic shows up in medicine.

  The first fuzzy relational database, FRDB, appeared in Maria Zemankova's 1983 dissertation. Other models followed: the Buckles-Petry model, the Prade-Testemale model, and the Umano-Fukami model. J. M. Medina, M. A. Vila, and others built a further model called GEFRED. Querying languages grew alongside them. SQLf came from P. Bosc and colleagues, and FSQL came from J. Galindo and colleagues. Both built structures like fuzzy conditions, comparators, constants, and thresholds directly into SQL statements.

  Computational theorist Bart Kosko argued, in an essay titled Fuzziness vs. Probability, that probability is actually a subtheory of fuzzy logic. He claimed that mutually exclusive outcomes in probability are just a special case of fuzzy logic's graded, non-exclusive membership. Kosko even derived Bayes' theorem from the concept of fuzzy subsethood. Lotfi Zadeh disagreed with that framing. He fuzzified probability itself into what he called fuzzy probability, and generalized the idea further into possibility theory. By 2016, the IEEE had turned decades of this formal work into its own Fuzzy Markup Language. That standard, built on XML, lets engineers describe a fuzzy system in a form any machine can read.