(Editor’s Intro) Our guest blogger for this entry is Larry Supina, who is the Manager of Integrated Safety Management and High Reliability Operations at the USDOE Pantex Plant. Larry has 20+ years of experience at Pantex, Masters Degree in Public Administration, Six Sigma Black Belt, and is certified in Human Performance Improvement. A couple of years ago Larry drew upon his expertise as a former meteorologist to develop a unique leading indicator concept for Pantex. Of particular note is that in this approach, the probabilistic challenges associated with prediction are set aside and the focus shifts to using indicators to promote new conversations that spark creative thinking about prevention. This further cultivates the prevention focused culture present today at Pantex.
Our thanks to Larry for sharing these Pantex insights!
Managers of modern organizations constantly find themselves in the position of responding to a series of unfortunate events, be it a rash of vehicle accidents, a spike in first aid cases, or a spate of unexpected operational delays. The resulting conversations are surprisingly predictable: “What do we do?” followed by “Why didn’t we see this coming?” The discussion then delves into developing and implementing a set of leading indicators so the management team can clearly see these types of events coming in the future. Diligent staff members head out to develop a leading indicator based on lagging data only to see the pattern repeat itself, again and again and again.
The Pantex Work Environment Forecast was born out of the frustration of trying to develop leading indicators for organizational events which inherently defy the development of such indicators. If you could accurately predict the next fatality at your site it is then a safe bet you would, and furthermore prevent it from occurring in the first place. This is not likely to happen as the variables within organizations, individuals and site conditions are simply too numerous, diverse and dynamic to realistically account for every scenario in a predictive manner.
Part of the problem, as already stated, is the vast amount of variable data which comes together at a single point in time to create an unwanted event. Past attempts in the safety discipline have been aimed at reducing the amount of data, to hopefully find the key data points which proved to have predictive value. On a small scale this can sometimes be accomplished, Job Task Analysis is a good example. However, this approach fails when it is scaled up beyond anything more than simple tasks or a series of tasks, as the analysis needs to bring in more and more data and with it, more variables.
Weather forecasting has many of the same problems: predicting exactly when and how much it will rain or snow in your backyard is amazingly difficult. Predicting that within a regional area some percentage of that area will receive a range of precipitation is manageable. This level of prediction, while not as specific as we may like, is close enough to give us the ability to plan activities and engage necessary precautions effectively. This is the direction that the Work Environment Forecast takes as a predictive management tool for minimizing unwanted events at a facility.
Before computers and satellites, weather forecasting depended heavily on historical models. These models still provide a baseline to compare computer models against. If a given computer model predicts snow in Houston on July 4th, the historical model provides confidence to the meteorologists that it’s time to re-boot the computer, not call out the non-existent snow plows in Houston. The historical model is based solely on past observations of weather behavior at a particular site, but more importantly for the application to site safety, the data is correlated to a calendar date. Conversely, the generally accepted method in use today for displaying safety related events is on a graph usually covering the last twelve months. For example, first aid cases over the past year which shows trends across consecutive time.
The genesis of the Work Environment Forecast was to look at facility specific data across periodic time frames. A bouncing trend-line across the last twelve months does little to encourage managers or workers to alter an existing behavior or start a new one. However, providing a forecast 30 days before the annual first snow of the season which demonstrates how for the last three years the first snow storm has resulted in freeze damaged pipes, hand injuries, and a high rate of vacation usage can result in behavior changes. This information is credible enough and time sensitive enough to provide reasonable people the stimulus to take actions and invoke precautions.
If organizations were found to have repeatable and predictable patterns of behavior at this level, then a much more precise level of management action could be taken by the managers and employees of a site to significantly reduce unwanted events. Our experience to date at Pantex is indeed that our facility/organization has inherent seasons driven by a host of factors, such as seasonal weather patterns, fiscal year budget cycles, employee sick leave and vacation use patterns, school year activities, sports activities and maintenance schedules. Instead of year after year responding to unwanted events by saying, “we should have seen that coming,” the Plant now receives a monthly forecast usually two weeks prior to the start of a new month outlining the changes the Plant will see during the upcoming month which provides specific actions and precautions to be taken beforehand.
Benefits of Forecasting
For Pantex, the benefits of Work Environment Forecasting (WEF) have far exceeded early assumptions. In the initial implementation the forecast was based on historical weather conditions and scheduled plant activities. The forum for presentation was and still is the monthly President’s Safety Council meeting which includes plant management, union representation and representatives from employee safety organizations such as Behavior Based Safety (BBS) committees and Voluntary Protection Program (VPP) committees. The forecast created an opportunity for open communications between management and these groups focused on proactively implementing safety. There is less reason for finger-pointing and more reason for collaboration when attempting to deal with upcoming conditions. An example from the first attempts at work environment forecasts which makes this point comes from the May 2006 WEF.
“May is a month of transition, snow is possible into the first week and summer heat will occur by month’s end.”
This simple sentence created a nearly half-hour dialogue between all the representatives in attendance. Plans were made to ensure adequate hydration of maintenance employees. An unknown employee grievance concerning lack of an adequate stock of a powdered drink mix during the summer of 2005 was aired and resolved. A proactive review of critical cooling systems was initiated. But more importantly, the entire conversation and ensuing action plans were undertaken in a spirit of teaming for the good of all.
Requests for copies of the Work Environment Forecast were received from all over Pantex. Managers and employees were seizing on this information for use in safety meetings, planning activities and as a catalyst for engaging employees in positive and proactive discussions about working safely. It was evident early on that the organization was not only receptive to this information, it was thirsting for it.
The second stage of development began looking at the organization’s historical data to determine if any patterns could be useful for prediction purposes. The first set of data utilized was the Plant Shift Superintendent’s log (PSS) kept by the Operations Center. The thought behind using this data was simple, these are the events the organization has a need and/or interest in reporting and logging. If the organization found these events of interest, it should stand to reason it would find any correlation of events and possible prediction interesting as well.
The Work Environment Forecast looks at employee leave usage over the past three years to provide an insight into expected behavior in the forecasted month. Holidays and weekends are not included in the data as these days vary significantly from the normal work-week pattern. Pantex has two distinct leave patterns, the school-season leave pattern and the summer pattern. Patterns which have also been identified and used for prediction include staff availability for the forecasted month, the spring and fall allergy seasons, expected number of no-shows due to snow storms, and expected impact created by policy changes.
Injuries are correlated across the last three years by month, which again provides considerably more insight than a running trend-line. Even the types of injuries the Plant experiences have seasons, and by alerting staff to an upcoming season we have been successful at mitigating hazards ahead of time and reducing the number of events.
Notable Findings to Date
The Three Day Cycle – One of the first correlations which intrigued us occurs regularly in November. Excessively high winds return to the Texas Panhandle in November which results in numerous alarms, equipment failures and injuries. Our first step was to increase the Plant’s awareness of how excessive the winds are by correlating them to tropical storm and hurricane strength winds. Living in the panhandle it is not uncommon to hear a weather forecast calling for gusty winds exceeding 40 to 50 mph; actually it is a daily fact of life from November to May. By re-stating we would experience 20+ days of tropical storm force winds and likely see 2-3 days of category 1 hurricane force winds in an upcoming month, and providing this information several weeks before the onset we noticed a decisive change in business. Loose items around the plant were secured, door-stops were replaced or repaired, and employees received safety briefings on how to protect their hands and eyes.
However we also noticed another pattern in the Co-Incidence Chart which took us awhile to fully understand. While wind-speeds and alarms were very tightly coupled, and equipment failures only lagged by a day, injuries, spills, process stops, and vehicle damage all increased three days after a major wind event. Which, we finally determined, is about how long it takes for a sinus infection to appear after a dust storm. Employees were losing focus and becoming distracted due to illness and the use of over-the-counter medications.
Vehicle Damage Twin Peaks – When an organization is trending data, such as vehicle damage, increases will be noted and campaigns will be started to gain employees’ attention. The increase will peak and as rates come down credit will be given to whatever campaign was put in place. Later the process will repeat itself, again and again. Using historical modeling, we noticed vehicle damage increases at the same general points in time each year. More specifically, when the weather is very hot and when the weather is very cold. These are the most optimum times from a human behavior point of view to take a little more risk and park a vehicle closer to a work-site, to try and get around an obstacle without getting out and looking, in general to stay inside the vehicle with the heater or air-conditioner running rather than suffer the elements.
Christmas in September – In August of 2009 while developing the September forecast an unusual co-incidence became apparent. Every September hand injuries came down from August highs but slips, trips and falls increased. One other single data point led us to the cause. Each September we recorded one injury to a custodian handling trash. At first tying these events together was proving difficult until we started making a list of things that happen at the Plant just in September. Fiscal year-end at government sites always sees a sudden increase in the purchase of supplies which have to be delivered quickly to make the close-out date. New equipment and supplies and old equipment vie for space and inevitably some materials are temporarily located in hallways, office floors and such until they can be disposed of. The trips and falls increase in September was a result of items being temporarily located in normally open walking spaces.
Personnel eager to get supplies stored and equipment replaced cleaned out spaces causing trash receptacles to be over-loaded or loaded with sharp objects resulting in the one custodian injury every year. Of course after the one custodian was injured, a notice was put out to the Plant. Every year we found a history of the Plant notifying employees to not overload trash cans and to safely dispose of sharp objects. Now we bring this to the employees’ attention in August before the opportunity to create a fall or cut hazard happens.
Hand Injury Season – Hand injuries, while they occur all year long, drop in frequency starting in December and stay low until May. Then the trend steadily increases each month until a peak is achieved in August. These events have little in common other than causing injury to an employee’s hand. Our speculation is cold weather creates a higher awareness of what your hands are doing and a higher chance gloves will be worn, providing some level of protection. Once we noticed this trend establishing itself again in FY 2009, a concerted campaign was aimed at having every employee watch the hands of everyone they came in contact with. This year our hand injuries peaked in July so we can with some confidence report at least 4 to 5 hand injuries were prevented.
Wind Injury Season – The wind blows in Amarillo, just about every day. However, May through October the winds stay in a more reasonable 20 mph range. Gusts beyond 30 mph during this time-frame are associated with severe thunderstorms which serve to heighten everyone’s awareness. Our injury data showed us that November through April was the significant time to attack wind-related injuries. The habits of putting a foot out to hold a door open and wrapping fingers around a door edge instead of using the handle get set during the calmer months, then when the southwest winds howl at hurricane strength unassociated with any storm, these habits turn into lacerations and fractures when doors caught by 60 mph winds show little mercy. The Work Environment Forecast now predicts these seasons and enables employees to focus on changing the specific habits which have been putting limbs in harm’s way.
Bites in the Fall – Each spring safety notices, articles and briefings comment on being careful to avoid animal and insect bites. This seems logical, as the weather warms up animals begin to move around, people start to head outside and hazardous encounters are sure to occur, right? Actually what we have found with the Work Environment Forecast is the greatest threat for bites and stings occurs as fall sets in. Animals and insects which have happily lived outside all year begin to move indoors to stay warm. This understanding has allowed Pantex to more accurately target the message to the hazard.
Perhaps the greatest accomplishment of the Work Environment Forecast is the simple act of having personnel think about what is going to happen in the workplace next month instead of waiting until it happens. Air conditioners are checked before it gets hot, heaters are checked before it gets cold, supplies are ordered in time to meet needs, safety briefings are relevant and specific, proactive conversations occur that result in preparing equipment, employees and processes instead of reactive conversations about why something happened. It is empowering to be prepared, and the culture of the Plant has shifted from the inherently negative position of responding to events to a much more positive and decidedly more effective position of actively applying the warnings provided by the Work Environment Forecast.
In short, lagging data only provides a forum for fault finding and morale destroying conversations whereas active forecasting, even if it is only partially correct, provides a forum for positive conversations, positive actions and in the event of failures it even provides a more positive methods for learning the lesson as the conversations are not centered so much on fault finding as on seeking better prediction methods.