<p>This paper proposes a torque estimation method using a Fuzzy Neural Network (FNN) for cordless impact wrenches equipped with oil-pulse mechanisms. While widely used in industrial fastening applications, such tools typically lack real-time torque output information, and their torque delivery is affected by tool aging and power variations. To address this problem, this paper proposes a sensorless method for estimating fastening torque using a data-driven FNN. The FNN consists of four inputs: voltage, motor current, motor speed, and impact duration. These variables are selected because they are available from the internal states of the tool without external sensors, thereby reducing implementation costs. This paper optimizes the FNN through supervised training data collected from experiments with new and aged tool mechanisms. The FNN characterized with structure and parameter learning is employed. Learning of the FNN starts from an empty structure, and new rules are online generated to properly cover input samples and build a compact network. The optimized FNN is implemented in the microcontroller unit (MCU) originally embedded in the tool to perform real-time estimation. That is, the proposed GNN-based estimation method does not require additional hardware implementation cost. Experimental results show that estimations of the FNN model meet industry-accepted torque tolerance and demonstrate good adaptability to tool aging and battery charge variation for practical feasibility.</p>