Multiple sensors provide insightful information for machine condition monitoring (MCM), and one mainstream method to comprehensively analyze high-dimensional sensor data is to construct a composite health index. Considering the difficulty in obtaining degraded data in practical production, this article proposes an optimal CHI based on control chart performance optimization for online monitoring. The optimal CHI minimizes average run length (ARL) of the control chart when a machine is in degraded conditions, facilitating the prompt detection of first prediction time (FPT). The effectiveness of the proposed method is validated by a turbofan engine dataset. Results show that the optimal CHI can not only detect the FPT in an early degraded stage, but also reveal informative sensors related to degraded module.