A simulated system has certain advantages over a physical one. It can be cheaper to implement, destroyed with little to no consequences and more generally toyed with basically no cost. To test fault detection algorithms, there is a need for the simulation of a faulty IoT network. In this paper, we present a platform for simulating IoT exchanges, where sensors can be real or virtual, only editors or both editors and consumers. The platform allows defining a wide variety of simulation parameters such as the number of sensors to instantiate, each sensor settings, perturbations to inject and stop conditions. We use the RabbitMQ broker for message exchanging and the Open Data Format for standardizing messages format. We detail the global architecture and present two case studies to show how the platform works. We build a "smart" sensor that retrieves measures from other sensors and uses them to predict the temperature value based on an artificial intelligence model and show that any model can be instantiated as a smart sensor.