Guest blog by Coen van Gulijk, professor of railway safety and risk at the University of Huddersfield.
Bowties are an effective visual representation of the risk space that you want to control. Bowties connect hazards, threats, consequences and most importantly controls in a clear interface. With the new BowTieServer, several functions can be added to bowties that provide you with relevant information about the quality and status of barriers: AuditXP, IncidentXP and Investigator 3 are examples of this. These products pave the way towards extensive data-integration for safety management systems: Big Data in your Bowtie. At the University of Huddersfield, we are working on adding even more data to the bowtie.
Big Data Risk Analysis
At the University of Huddersfield, we believe that the future of safety management is digital. Now, it is easy to claim that the future is digital as just about any industry is digitizing but it is not so easy to actually make it happen. We started our journey from the definition below.
A BDRA management system is defined an enterprise safety management system that:
- extracts information from mixed data sources;
- processes them quickly to infer and present relevant safety management information;
- combines applications to collectively provide sensible interpretation;
- uses online interfaces to connect the right people at the right time.
In order to:
- provide decision support for safety and risk management.
This is not about Captain Picard’s imaginary starship but it is IT jargon. I am afraid there is no way around the ingress of IT technology in our field and will have to get used to some of its jargon. Safety experts may have to become a bit more data-savvy. Despite that, I think our definition has parts that are well-trodden for safety experts users: decision support, safety information and a clear visual interface in the form of a bowtie. In many ways, BowTieServer is an example of a BDRA system.
But what is it that state-of-the-art IT systems and data have to offer? I’d like to offer a number of views. First, there is an opportunity for lean and efficient management delivery by connecting pads and phones with a central repository for a safety management system; something that we keep seeing more and more of in the field. Second, with that connectivity we can strive for real-time risk management, not just for calamities but also for more mundane risk controls so we could detect rapidly deteriorating barriers quickly. Third, with the help of AI based systems we can increase staff inclusion and engagement for instance with interactive event reporting where the system asks for additional information about barriers based on information provided. And finally, if we can integrate our incident data with alternative data sources, we have a richer historic database to work from. BowTieServer does that in the sense that it combines incident data, investigations and audits but more integration is possible: near miss reporting systems, maintenance reports, fault detection systems can all enrich our view of the status of a barrier. All these fields are being developed; our work focuses on the latter.
Natural Language Processing
A key step for integration and interpretation of data is Natural Language Processing or NLP. Often the most relevant safety information is captured in text, which is easily understood by humans but it takes some effort for computers to get their silicon heads around. Fortunately, text analysis is an active research field in IT and quite a few technologies are out there. The figure below is a part of a word-cloud of 500 near reporting incidents we published in earlier work and it identifies a risk that is described in them without reading them all in detail: level crossing risk. That same word-cloud technique enables us to find all text records that mention particular risks or barriers. That forms the basis for automatically linking any safety-relevant text document to a bowtie. The accuracy of automatic linking varies but that does not keep us from trying to link reports from legacy systems, near miss reporting systems, maintenance reports, e-mails and standards.
Figure 1: Wordcloud of 500 near reporting incidents
Linkage of Near Miss Reports
With NLP as a key enabler, we have linked text-based near-miss reports to an existing bowtie and interfaced with BowTieServer. Figure 2 shows a small part of a BowTieServer interface populated with incident data from an alternative data source. The example here is hardly Big Data but the method and approach are scalable.
Figure 2: Bowtie with linked text-based near-miss report data
The example is not Big Data just yet, but the method and approach are scalable. In addition to that, we are extracting safety KPI’s from data from the railway signaling system and on-train data recorders which are formidable data sets. Thus we are mixing different data sources to serve a single bowtie for efficient risk control. We believe that these techniques pave the way to a whole new level of safety control and a central role for bowties as integration platforms.
Do you want to learn more about big data and bowties? At the complimentary Risk Management in Rail event in York June 5th, you will have the opportunity to explore this topic further with Prof. Coen van Gulijk. For more information, click here.