An Empirical Study on Accessibility Improvement for Non-experts Using a Natural Language - Based PLC Ladder Logic Generation Frame work
This study presents and empirically evaluates a vendor-independent framework that converts natural-language instructions into IEC 61131-3-compliant PLC ladder logic and deploys the generated code to Siemens, Mitsubishi, Rockwell, and LS Electric hardware. The study addresses two persistent barriers in smart manufacturing: shortages of skilled PLC engineers and vendor lock-in across heterogeneous development environments. The experiment involved 60 participants(20 experts and 40 students) performing basic, intermediate, and advanced tasks. Experts completed the Baseline condition only, whereas students completed both Baseline and Proposed conditions, yielding 300 participant–task–condition logs and 1,500 metric data points. For the 40-student paired dataset(df=39), paired-samples t-tests, Cohen's d, and 95% confidence intervals were computed. In the advanced task, student completion time decreased from 184.51 min to 58.41 min, while the expert-relative gap narrowed from 130.13 min to 4.03 min. Code optimality improved from 60.88% to 92.31%, logic errors decreased by 83.89%, safety-rule violations decreased by 89.71%, and SUS increased from 43.41 to 83.81. All task–metric combinations were significant(p<0.001), and vendor validation succeeded in all 60 scenarios with mean conversion times of 1.9–2.6 s. These findings indicate that a constrained, human-verified natural-language interface can substantially improve non-expert PLC task performance within the evaluated task range while preserving cross-vendor deployability