It is essential to avoid deadlock and blockage in the operation of an automated manufacturing system (AMS) since they cripple a part of the system, if not the whole, and then deteriorate system performance. Though a large number of approaches are developed to resolve this issue, few of them are directly applicable to AMSs with failure-prone resources. This study focuses on the supervision for deadlock avoidance and blockage control in a class of AMSs with multiple unreliable resources, where each resource is a workstation. The authors hope that the system, under the control of a supervisor, can at all time automatically continue producing parts whose processing routes do not contain any of the currently failed workstations. Inspired by the widely known Banker's algorithm and a resource order policy, this study presents a robust supervisory controller that satisfies the above desired requirement. All the evidence shows that authors' policy performs better than the original one in terms of permissiveness. Furthermore, an example AMS from existing literature is used to illustrate the effectiveness of authors' proposed method.