The Intelligent Instrument course requires experimental cases that integrate sensing, data acquisition, signal processing, and intelligent analysis into a complete engineering workflow. This paper develops a squirrel-cage induction motor rotor fault diagnosis case for this purpose and proposes a lightweight wavelet-impulse convolutional neural network for seven-class diagnosis. A differentiable impulse separation module decomposes the measured phase-voltage signal into complementary background and impulse-related branches, followed by a Morlet-initialized learnable filter bank for adaptive feature extraction. Experiments show that the model achieves 99.96% accuracy under clean conditions and 89.17% at 5 dB Gaussian noise, outperforming the best comparison model by 9.66 percentage points. With only 11.43 K parameters and a 0.05 MB model size, the case provides a compact example connecting physical sensing, signal processing, and AI-based condition identification for Intelligent Instrument teaching.

Squirrel-cage induction motor, differentiable impulse separation, learnable wavelet filter bank, lightweight neural network