Engineering Consultants

Use of Predictive Analytics in EHS

Summary
These days EHS (Environment, Health, Safety) performance of any operation or site can be “Predicted” with reasonable accuracy by using the field of modern computing known as Predictive Analytics. Whenever there is too much information (Data) which a human mind can’t comprehend or make a sense of, computers with complex calculations (Algorithms) can still make sense of all that data to see emerging patterns (Pattern Recognition) and Predict the likelihood of any event. This is a relatively new field of computing and is producing amazing results in nearly all fields of life. This article discusses the use of Predictive Analytics in the field of EHS.

Use of Predictive Analytics in EHS

Major industrial accidents are a stark reminder of how a simple mistake can start a chain of events which ends in several deaths and hundreds of millions of dollars in financial loss. The recent incidents in a refinery in the Middle East and power plant in India were such reminders.

It is a proven fact that major incidents don’t happen overnight, there always are telltale signs which are ignored, overlooked or hidden.

We are all familiar with the classic incident pyramid which has been modified over the years but the crux of it stayed unchanged i.e. 1-10-30-600 (each serious injury or death is preceded by approximately 10 minor injuries or 30 property damages or 600 nearmisses).

Use of Predictive Analytics in EHS

Incident Pyramid

In traditional approach, if a facility had approx. 600 nearmisses, being the leading indicator, it could predict that a serious injury was just round the corner, but that is where the information would stop. The Where & When was never known.

If you have been in the field of EHS (Environment, Health, Safety & Sustainability) long enough, you will agree that there are basically 3 types of organization cultures:

Culture-1: HIDE
The management is only concerned about good EHS statistics and hides nearmisses & incidents and is not willing to invest in the required resources. Such organizations develop a culture of hiding nearmisses, incidents & problems as much as they can till an incident so big happens that it can’t be hidden e.g. explosion resulting in several deaths. This is the most tragic way of learning from the past.

Culture-2: DISCOURAGE
The management wants to have a good EHS culture but discourages reporting nearmisses & small incidents without realizing the detrimental effect it can have and then end up with a big incident or death.

Culture-3: ENCOURAGE
The management genuinely wants to have a good EHS culture and encourages employees to report all nearmisses & incidents. Such organizations have exemplary EHS record and rarely end up with a major incident.

Predictive Analytics is a tool in the hands of willing management which supports & promotes the Culture-3 mentioned above. So what really is Predictive Analytics?

A lot of EHS professionals are either unaware or unsure of what Predictive Analytics is and how it can benefit the CPI (Chemical Process Industry). To top it all, there are so many confusing terms being thrown around randomly that no one, other than the subject experts, really seems to know what to make out of those terms e.g. big data, data analytics, machine learning, predictive analytics, artificial intelligence, pattern recognition, cloud computing and so on.

As an EHS professional, one of the most important things that keep you busy is trying to ensure a safe future and avoid accidents. This is one of the most difficult challenge and almost impossible to predict with a reasonable degree of accuracy up until now.

With Predictive Analytics the Where & When can be known with a reasonable degree of certainty, in some cases with a certainty of upto 97%. The use of Predictive Analytics is reshaping things in the EHS field for those organizations that have started using it.

predictive analytics

The purpose of this article is to consolidate the relevant information so that if an EHS professional wanted to implement a Predictive Analytics program, he would know why & how to do it.

With the advent of data analytics and reliance on leading indicators vs lagging indicators, highly accurate predictions can be made.

To refresh the memory:

Lagging Indicators rely on something AFTER it has happened in the past and use it as a learning experience to prevent future incidents e.g. findings from an EHS accident investigation report.

Leading Indicators use the information BEFORE any incident has happened e.g. EHS inspection reports, nearmiss reports etc. which can show a trend of complacency which might lead to an accident.

To fully understand the role of Predictive Analytics, we have to understand the decision making process of a human mind.

We can’t make good decisions unless we understand the problem (data) in front of us. The human mind is inherently limited in handling data beyond a certain size and seeing patterns in large data (commonly known as Big Data). Without seeing any pattern in data, we can’t make smart decisions. So how big is big data?

Big Data is a general term without any exact definition e.g. it would have been easy to say, in computer processing & memory terminology, anything above 32 GB (gigabytes) is big data but that is not the case. Human minds don’t work with the units of gigabytes or megabytes. So any information or data large enough for the human mind to handle and process, be it a genius, would be big data.

To understand the concept of big data and the use of computers for handling that, we have to go in the past, in the early 1940s when the computer as we know it now was invented by a genius Alan Turing. He invented the computer to decrypt the code of Enigma machine which the Germans were using to send encrypted messages across enemy lines causing heavy losses to the British. The British despite having some of the best minds couldn’t decrypt those coded messages because the simplest Enigma machine was capable of generating 15,000,000,000,000,000,000 combinations for every message that it sent. This was humanly impossible to decrypt. It was Big Data! You can watch the movie “The Imitation Game” where Alan Turing, played by Benedict Cumberbatch, breaks the code of Enigma Machine with the first modern computer.

Use of Predictive Analytics in EHS

Enigma Machine

Now that you know what big data and why it is humanly impossible to analyze & understand it, we need computers to do it. The computers use Machine Learning based on Artificial Intelligence to process big data and try to “see” patterns which the human mind can’t. This data can be related to any field not only EHS e.g. Finance, Health Industry etc. It is sometimes so huge and coming from different sites & organizations that it can’t be located at any one site but rather housed in data houses in interconnected remote servers (commonly known as cloud computing).

So how does all of this relate to EHS? Imagine for a moment that the EHS suggestions, nearmisses & EHS inspection reports from different departments of several similar organizations were gathered for years. These documents would run into thousands. If all of these were compiled in a logical manner and then a computer program sifted through that data trying to look for patterns of incidents to predict the likelihood in the future, that would be Predictive Analytics. It is as if thousands of geniuses were working on a problem at the same time. The more data the computer has, the more accurate the prediction.

Many organizations that successfully implement EHS programs and have low incident rates will run out of incident data to learn from. Such organizations need leading indicators & Predictive Analytics to help them stay safe.

So in terms of EHS, what is big data. Simply put all the EHS reports, gathered over years and from various chemical plants will form big data.

Predictive analytics relies on the following main processes:

Analytics:
1. Statistics
2. Data mining
3. Text mining
4. Machine learning
a. Artificial Intelligence
b. Visualization
5. Reporting

Decision Optimization
1. Scoring engine
2. Rules engine
3. Recommendation engine
4. Optimization engine

One of the most commonly used method for Predictive Analytics is Regression modeling which determines the relationship between known & unknown variables. There are several regression techniques ranging from the simplest Linear Regression to the complex Elasticnet regression. There are other techniques also like Neural Networking, Support Vector Machines etc.

Exactly how the predictive analytics models work and their mathematical reasoning is beyond the scope of this article, suffice it to say that all such models use advanced statistical analysis methods and based on the past data try to predict what the future outcome will be for a particular set of known variables.

The outcome we get by using Predictive Analytics is that if a safety incident happened in the past when a particular combination of parameters & conditions was present, then it is safe to assume that such an incident will happen again, if the same or a similar combination of parameters & conditions happens.

Many organizations that successfully implement EHS programs and have low incident rates will run out of incident data to learn from. Such organizations need leading indicators & Predictive Analytics to help them stay safe.

As per the Pareto principle (also known as the 80-20 law i.e. 80% of the effects come from 20% of the causes), it would be safe to assume that more than 80% of the organizations still use the lagging indicators which can only work AFTER the fact. As shown above, this approach worked successfully upto a few years back but if you want to stay ahead of the incidents, Predictive Analytics is a very strong tool to have on your side.

Remember, a tool is only as good as the person using it, so Predictive Analytics can’t solve all your problems and if used incorrectly, you will only get disappointment.

A successful Predictive Analytics program hinges on the following points:

1. Goals & Objectives: Have SMART (Specific Measurable Achievable Realistic Time bound) objectives that you want to achieve e.g. a 10% drop in nearmisses in 2 years once the program is operational.
2. Accurate data: This is the lynchpin, the bedrock, the core, the foundation of your Predictive Analytics program and can’t be emphasized enough. Spend as much time as required to make sure that your data is rock solid. Discard any & everything which you are unsure about e.g. fake suggestions & reports which have only been done to meet some deadlines. Remember the rule of GIGO (Garbage In- Garbage Out) so unless your data is spot on, you will only get disappointment in implementing a Predictive Analytics program.
3. Plan everything: Most Predictive Analytics programs take upto 80% of their time in planning & data validation so don’t be afraid to spend most of the time on planning phase, the rest would be simple.
4. Start small: If you have never done this sort of work before, start small and build upon your successes. Diving into the deep end of the pool when you are still a novice swimmer is not a very smart strategy.
5. Team work: Predictive Analytics requires a good team to be a success. Choose your team sensibly which should have a representative expert from all required fields e.g. EHS, IT.

In the nutshell, PA (Predictive Analytics) is a tool, among other tools, which will help you make informed decisions about operating your plant in a safe manner which also is the most economical manner.

References:
1. Shultz, G., Advanced and predictive analytics in safety
2. Robert L. Mitchell, 12 predictive analytics screw ups
3. Brian Gilbert, Using data analytics to improve EHS and sustainability performance
4. Elliot Laratonda, Predictive Analytics Using Leading Indicators to Prevent the Next Injury
5. Sunil Ray, Comprehensive guide regression
6. OSHA, USA