Built-in test (BIT) technology is gradually employed in predictive maintenance of equipment state. However, the accurate signal state judgment of conventional BIT (CBIT) is affected by high false alarm rate (FAR). To alleviate this issue, an IFCN-BIASN based BIT signal state recognition approach for heavy-duty gas turbine controller is proposed. Improved fully convolutional network (IFCN) and bio-inspired adaptive synaptogenesis network (BIASN) are integrated in the proposed approach. The impacts of signal noise and intermittent faults on BIT results are effectively reduced. At first, false alarms and missed alarms are diluted by IFCN. Then, the remaining isolated false alarms and missed alarms are efficiently eliminated by BIASN. Finally, the proposed approach is tested on four categories of BIT signal datasets. The ablation experiments validate that the improvements of the proposed approach have played their due roles. Moreover, the proposed approach is able to achieve more accurate recognition results compared with the CBIT and several available signal state recognition approaches.