This paper proposes a method to improve Knowledge Base Expert System (KBES) of Some Large Crane, grounded on Expert Rules. On detecting failures of the Large Crane, the improved system can diagnose almost all possible reasons for failures, and measure the occurrence probability of each. The diagnosis process is supported by reasoning, according to Bayesian Network, constructed by Expert Experience. Another feature of this system is adopting a method to initialize Bayesian Network and learn Bayesian Network CPT. By comparison, applications of conventional and improved expert system to failure diagnosis presented in this paper, illustrate that the latter can identify the cause of failure more promptly and accurately.