It's correct that knowledge based Expert systems have fallen out of fashion. They were intensive researched in the 1970s and 1980s. The new, more exciting topic, is called data-driven and machine learning approach. This kind of turnaround can be traced back in the literature. Since the 1990s lots of papers were created about data-mining and only a few literature was written about classical handcrafted expert systems.
Are expert system still in use?
The dinosaur from the 1980s was CLIPS. The sourcecode is available in the internet and a new Java version was programmed called Jess, but in general it is no longer a topic in the literature. Instead of focus on expert systems, many researchers have selected semantic web and OWL as their new home base. Some extensions in the OWL syntax have to do with creating rules similar to what was researched in the 1980s under the term expert system.
The shared similarity between expert systems and machine learning is, that both are transfering input into output signals in a control systems. A plant has input values which are measured by sensors and it has also output signals which are valves. The aim of the system in-between is to realize a certain policy. Expert systems are doing so with symbolic stored knowledge, while machine learning is using statistical patterns for the signal mapping.
According to the given literature and the amount of failed automation projects in the past, it can be sure, that the experts doesn't know how to realize the policy in optimal way. In the 1970s and 80s, expert systems were used, and later the machine learning concept. With both it is possible to store domain-knowledge in a machine readable way and perhaps the ideal plant control would be a hybrid of both.