Rotating machinery equipment fault detection and identification has always been a research hotspot. Aiming at the deficiency of current fault feature extraction and diagnosis methods, this study proposes a method based on improved complete ensemble empirical mode decomposition (ICEEMD) and adaptive whale optimization algorithm (AWOA) optimized extreme learning machine (ELM). The generation of pseudomodality can be avoided by ICEEMD during the decomposition process, and the residual noise in the mode is small. The extracted fault information is more accurate. ICEEMD is used to decompose the collected signals into intrinsic mode function (IMF). Through analyzing Spearman rank correlation coefficient (SRCC) among IMFs of rolling bearings in different fault states, the conclusion is that the IMF should be screened out when its SRCC is larger than 002. The hybrid entropy (HE) of the screened IMF is further calculated as feature vectors. Compared with other bionic algorithms, the whale optimization algorithm (WOA) has advantages of fewer related parameters to be adjusted, faster convergence speed, and better stability. AWOA improves the convergence accuracy further through optimizing WOA′s local search mode by adaptive weight. Through AWOA optimizing the weight and threshold of ELM, the accuracy of fault diagnosis is improved. Comparison experiments show that AWOAELM has strong learning ability and higher accuracy of fault diagnosis. The AWOAELM method is applied to the fault diagnosis of ball bearings and outer rings of rolling bearings with different sizes. The accuracy of ball fault diagnosis is 995%, and the accuracy of external loop fault diagnosis is 100%.