This study investigates a system for collecting and analyzing defect information of stacked assemblies during manufacturing processes that utilize pneumatic actuators—specifically air cylinders—without relying on external sensors. The objective is to precisely analyze the pneumatic consumption patterns before and after the piston rod of an air cylinder reaches its target stroke and contacts the stacked assemblies. This approach enables the classification of defect types, such as cracks and missing layers, without the need for additional devices like vision sensors or load cells. The system was developed by collecting pneumatic data under both normal and defective conditions from an air cylinder installed on a turntable used for assembling circuit breaker trip units. After securing an operational margin through data analysis, pneumatic consumption patterns surrounding the moment of piston-to-stacked assembly contact were extracted and analyzed. By evaluating thrust variations caused by defects such as cracks or missing layers, the system can effectively identify and classify defect types in the stacked assemblies. Traditionally, signals such as current, voltage, pneumatic pressure, and temperature are monitored for predictive maintenance by assessing the condition of actuators like motors, heaters, or air cylinders. However, anomalies in these parameters do not always correspond to defects in the processed stacked assemblies. In contrast, this study focuses on directly capturing the influence of actual physical defects—such as missing layers, cracks, or assembly misalignments—on the behavior of the air cylinder. This enables accurate and sensorless quality control, representing a key differentiator of the proposed approach. Artificial intelligence models, including Random Forest and Convolutional Neural Networks (CNNs), were employed to learn the pneumatic consumption patterns and predict defects. The Random Forest model demonstrated strong performance in distinguishing between normal and defective conditions, while the CNN model showed high sensitivity in detecting specific defect patterns.