REAL

**REAL is a 32-bit single-precision floating-point data type** following the IEEE 754 standard. The default floating-point type for most PLC arithmetic — analog scaling, PID loops, engineering unit conversions, and any calculation involving fractions.

**Range:** approximately ±3.4 × 10³⁸ with about 7 significant decimal digits of precision.

**Memory:** exactly 4 bytes (32 bits): 1 sign bit, 8 exponent bits, 23 mantissa bits.

**Common uses:** - Scaled analog values in engineering units (°C, PSI, GPM, RPM) - PID loop variables (process variable, setpoint, control output) - Math involving fractions (averaging, conversion factors) - Sensor calibration coefficients - Recipe values that aren't whole numbers

**Syntax across major IDEs:** - Studio 5000 ST: `temperature : REAL := 0.0;` (Allen-Bradley's only floating type; LREAL is rare) - Siemens SCL: `temperature : REAL := 0.0;` - CODESYS / TwinCAT: `temperature : REAL := 0.0;`

**Critical pitfalls (the bugs that cost weeks):** - **Precision** — REAL has ~7 decimal digits. The integer 16,777,216 (2²⁴) is exactly representable, but 16,777,217 rounds to 16,777,216. If you're storing currency in cents, NEVER use REAL — use DINT. - **NEVER compare REALs with `=`** — use a tolerance: `IF ABS(a − b) < 0.001 THEN`. `0.1 + 0.2 ≠ 0.3` exactly in IEEE 754. - **NaN and Infinity** — division by zero produces ±Infinity; sqrt(-1) produces NaN. Both propagate silently and break downstream comparisons. Always check inputs. - **Endianness** — Modbus floats are notoriously inconsistent. Different vendors use different byte orderings (big-endian, little-endian, byte-swapped, word-swapped). When integrating, always test with a known value. - **Mixed types** — `myREAL := myDINT / 2;` gives integer division on most IDEs. Use `myREAL := DINT_TO_REAL(myDINT) / 2.0;`.

**Conversion:** `REAL_TO_DINT()` rounds toward zero; `REAL_TO_LREAL()` is lossless; `INT_TO_REAL()` is lossless for all 16-bit values.

**See also:** LREAL (64-bit double-precision), DINT (32-bit integer for currency/counts), TIME (time durations without REAL precision issues).