Model-based diagnosis is a new intelligent diagnosis technique which can overcome the shortcomings of traditional diagnostic methods. Hierarchical diagnosis is an important method for reducing the complexity of model-based diagnosis. In this paper, hierarchical description of the device to be diagnosed is proposed and the relationships between different abstract levels are pointed out. The soundness of hierarchical diagnosis is also proved: a diagnosis at the abstract level has a corresponding diagnosis at the detailed level. Furthermore, the incompleteness of hierarchical diagnosis is pointed out: a diagnosis at the detailed level may not have a corresponding diagnosis at its abstract level.

The methods for computing the kernel consistency-based diagnoses and the kernel abductive diagnoses are only suited for the situation when part of the fault behavioral modes of the components are known. The characterization of the kernel model-based diagnosis based on the general causal theory is proposed, which can breakthrough the limitation of the above methods when all behavioral modes of each component are known. Using this method, when observation subsets deduced logically are respectively assigned to be empty or the whole observation set, the kernel consistency-based diagnoses and the kernel abductive diagnoses can deal with all situations. The direct relationship between this diagnostic procedure and the prime implicants/implicates is proved, which links theoretical result with implementation.

最近，许多学者意识到了将因果关系这一概念应用于基于模型的诊断领域的重要性。然而，他们的研究只局限于简单因果理论。本文提出的广义因果理论包容了更多的信息。本文指出：广义因果理论的诊断空间小于等于相应简单因果理论的诊断空间。文中给出了当待诊断系统的模型为广义因果理论时的基于模型的诊断、基于模型的中心诊断等概念，论证了基于模型的中心诊断与本原蕴含/蕴含式的直接关系，从而将本文的理论结果与实现联系起来。本文进一步指出，对广义因果理论的基于一致性中心诊断和中心溯因诊断的刻划仅仅是本文所给出的刻划的两个特殊情形.

基于模型的诊断是一项新型的智能推理技术，它可克服传统诊断方法的缺点。分层诊断是减少基于模型诊断计算复杂性的重要方法。以往的分层诊断方法局限于使用单一、固定的分层描述，而同一个分层描述有时能提高诊断的效率，但有时会与非分层诊断效率相同。本文通过使用相应结构抽象树重新安排各层给定的分层描述来创建一个新的分层描述，提出一项新的动态分层诊断技术来提高诊断效率。

本文给出了基于一致性的最小正常诊断的概念，并将它与基于一致性的最小反常诊断加以区别。我们证明了：若将系统描述限制在故障理论或在故障理论中扩展有限的正常行为知识，那么使用最小正常诊断能够刻划基于一致性的诊断空间，即最小正常诊断假设成立。我们指出：使用最小正常诊断可以缩小只针对正常行为模型的最小反常诊断所产生的诊断空间，帮助我们找到真正的故障。文中最后还指出了最小正常诊断所适合的诊断任务及诊断领域。